arm_compute v19.11
diff --git a/src/graph/backends/CL/CLDeviceBackend.cpp b/src/graph/backends/CL/CLDeviceBackend.cpp
index 0666ec0..de31847 100644
--- a/src/graph/backends/CL/CLDeviceBackend.cpp
+++ b/src/graph/backends/CL/CLDeviceBackend.cpp
@@ -34,11 +34,13 @@
 #include "arm_compute/graph/backends/CL/CLSubTensorHandle.h"
 #include "arm_compute/graph/backends/CL/CLTensorHandle.h"
 
+#include "arm_compute/core/CL/CLCoreRuntimeContext.h"
 #include "arm_compute/core/TensorInfo.h"
 #include "arm_compute/runtime/BlobLifetimeManager.h"
 #include "arm_compute/runtime/CL/CLBufferAllocator.h"
-#include "arm_compute/runtime/CL/CLMemoryGroup.h"
 #include "arm_compute/runtime/CL/CLScheduler.h"
+#include "arm_compute/runtime/IWeightsManager.h"
+#include "arm_compute/runtime/MemoryGroup.h"
 #include "arm_compute/runtime/MemoryManagerOnDemand.h"
 #include "arm_compute/runtime/PoolManager.h"
 
@@ -90,9 +92,8 @@
 {
     // Setup Scheduler
     CLScheduler::get().default_init(&_tuner);
-
     // Create allocator with new context
-    _allocator = support::cpp14::make_unique<CLBufferAllocator>();
+    _allocator = support::cpp14::make_unique<CLBufferAllocator>(nullptr /* legacy path for CLCoreRuntimeContext */);
 }
 
 void CLDeviceBackend::release_backend_context(GraphContext &ctx)
@@ -132,11 +133,21 @@
         mm_ctx.target      = Target::CL;
         mm_ctx.intra_mm    = create_memory_manager(MemoryManagerAffinity::Buffer);
         mm_ctx.cross_mm    = create_memory_manager(MemoryManagerAffinity::Buffer);
-        mm_ctx.cross_group = std::make_shared<CLMemoryGroup>(mm_ctx.cross_mm);
+        mm_ctx.cross_group = std::make_shared<MemoryGroup>(mm_ctx.cross_mm);
         mm_ctx.allocator   = _allocator.get();
 
         ctx.insert_memory_management_ctx(std::move(mm_ctx));
     }
+
+    // Create function level weights manager
+    if(ctx.weights_management_ctx(Target::CL) == nullptr)
+    {
+        WeightsManagerContext wm_ctx;
+        wm_ctx.target = Target::CL;
+        wm_ctx.wm     = create_weights_manager();
+
+        ctx.insert_weights_management_ctx(std::move(wm_ctx));
+    }
 }
 
 bool CLDeviceBackend::is_backend_supported()
@@ -204,6 +215,12 @@
 
     return mm;
 }
+
+std::shared_ptr<arm_compute::IWeightsManager> CLDeviceBackend::create_weights_manager()
+{
+    auto weights_mgr = std::make_shared<IWeightsManager>();
+    return weights_mgr;
+}
 } // namespace backends
 } // namespace graph
 } // namespace arm_compute
diff --git a/src/graph/backends/CL/CLFunctionsFactory.cpp b/src/graph/backends/CL/CLFunctionsFactory.cpp
index 82b6dd6..57b48b0 100644
--- a/src/graph/backends/CL/CLFunctionsFactory.cpp
+++ b/src/graph/backends/CL/CLFunctionsFactory.cpp
@@ -25,6 +25,7 @@
 
 #include "arm_compute/core/utils/misc/Cast.h"
 #include "arm_compute/graph/Graph.h"
+#include "arm_compute/graph/GraphContext.h"
 #include "arm_compute/graph/backends/FunctionHelpers.h"
 #include "arm_compute/runtime/CL/CLFunctions.h"
 #include "arm_compute/runtime/CPP/CPPFunctions.h"
@@ -56,13 +57,6 @@
     using WinogradConvolutionLayer = CLWinogradConvolutionLayer;
 };
 
-/** Collection of CL depthwise convolution functions */
-struct CLDepthwiseConvolutionLayerFunctions
-{
-    using GenericDepthwiseConvolutionLayer   = CLDepthwiseConvolutionLayer;
-    using OptimizedDepthwiseConvolutionLayer = CLDepthwiseConvolutionLayer3x3;
-};
-
 /** Collection of CL element-wise functions */
 struct CLEltwiseFunctions
 {
@@ -249,7 +243,9 @@
         case NodeType::ConcatenateLayer:
             return detail::create_concatenate_layer<CLConcatenateLayer, CLTargetInfo>(*polymorphic_downcast<ConcatenateLayerNode *>(node));
         case NodeType::DepthwiseConvolutionLayer:
-            return detail::create_depthwise_convolution_layer<CLDepthwiseConvolutionLayerFunctions, CLTargetInfo>(*polymorphic_downcast<DepthwiseConvolutionLayerNode *>(node));
+            return detail::create_depthwise_convolution_layer<CLDepthwiseConvolutionLayer, CLTargetInfo>(*polymorphic_downcast<DepthwiseConvolutionLayerNode *>(node));
+        case NodeType::DequantizationLayer:
+            return detail::create_dequantization_layer<CLDequantizationLayer, CLTargetInfo>(*polymorphic_downcast<DequantizationLayerNode *>(node));
         case NodeType::DetectionOutputLayer:
             return detail::create_detection_output_layer<CPPDetectionOutputLayer, CLTargetInfo>(*polymorphic_downcast<DetectionOutputLayerNode *>(node));
         case NodeType::DetectionPostProcessLayer:
@@ -261,9 +257,9 @@
         case NodeType::FullyConnectedLayer:
             return detail::create_fully_connected_layer<CLFullyConnectedLayer, CLTargetInfo>(*polymorphic_downcast<FullyConnectedLayerNode *>(node), ctx);
         case NodeType::FusedConvolutionBatchNormalizationLayer:
-            return detail::create_fused_convolution_batch_normalization_layer<CLFusedLayerTypes, CLTargetInfo>(*polymorphic_downcast<FusedConvolutionBatchNormalizationNode *>(node));
+            return detail::create_fused_convolution_batch_normalization_layer<CLFusedLayerTypes, CLTargetInfo>(*polymorphic_downcast<FusedConvolutionBatchNormalizationNode *>(node), ctx);
         case NodeType::FusedDepthwiseConvolutionBatchNormalizationLayer:
-            return detail::create_fused_depthwise_convolution_batch_normalization_layer<CLFusedLayerTypes, CLTargetInfo>(*polymorphic_downcast<FusedDepthwiseConvolutionBatchNormalizationNode *>(node));
+            return detail::create_fused_depthwise_convolution_batch_normalization_layer<CLFusedLayerTypes, CLTargetInfo>(*polymorphic_downcast<FusedDepthwiseConvolutionBatchNormalizationNode *>(node), ctx);
         case NodeType::GenerateProposalsLayer:
             return detail::create_generate_proposals_layer<CLGenerateProposalsLayer, CLTargetInfo>(*polymorphic_downcast<GenerateProposalsLayerNode *>(node), ctx);
         case NodeType::NormalizationLayer:
diff --git a/src/graph/backends/CL/CLNodeValidator.cpp b/src/graph/backends/CL/CLNodeValidator.cpp
index 40ec508..8ca58bc 100644
--- a/src/graph/backends/CL/CLNodeValidator.cpp
+++ b/src/graph/backends/CL/CLNodeValidator.cpp
@@ -58,8 +58,9 @@
                    CLGEMMConvolutionLayer,
                    CLWinogradConvolutionLayer>(*polymorphic_downcast<ConvolutionLayerNode *>(node));
         case NodeType::DepthwiseConvolutionLayer:
-            return detail::validate_depthwise_convolution_layer<CLDepthwiseConvolutionLayer,
-                   CLDepthwiseConvolutionLayer3x3>(*polymorphic_downcast<DepthwiseConvolutionLayerNode *>(node));
+            return detail::validate_depthwise_convolution_layer<CLDepthwiseConvolutionLayer>(*polymorphic_downcast<DepthwiseConvolutionLayerNode *>(node));
+        case NodeType::DequantizationLayer:
+            return detail::validate_dequantization_layer<CLDequantizationLayer>(*polymorphic_downcast<DequantizationLayerNode *>(node));
         case NodeType::DetectionOutputLayer:
             return detail::validate_detection_output_layer<CPPDetectionOutputLayer>(*polymorphic_downcast<DetectionOutputLayerNode *>(node));
         case NodeType::DetectionPostProcessLayer:
diff --git a/src/graph/backends/CL/CLTensorHandle.cpp b/src/graph/backends/CL/CLTensorHandle.cpp
index 219d9d0..891c784 100644
--- a/src/graph/backends/CL/CLTensorHandle.cpp
+++ b/src/graph/backends/CL/CLTensorHandle.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2018 ARM Limited.
+ * Copyright (c) 2018-2019 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -23,8 +23,7 @@
  */
 #include "arm_compute/graph/backends/CL/CLTensorHandle.h"
 
-#include "arm_compute/core/utils/misc/Cast.h"
-#include "arm_compute/runtime/CL/CLMemoryGroup.h"
+#include "arm_compute/runtime/IMemoryGroup.h"
 
 namespace arm_compute
 {
@@ -52,8 +51,7 @@
 {
     if(mg != nullptr)
     {
-        auto *cl_mg = arm_compute::utils::cast::polymorphic_downcast<CLMemoryGroup *>(mg);
-        cl_mg->manage(&_tensor);
+        mg->manage(&_tensor);
     }
 }