Set up the framework to choose the default LWS

Resolve COMPMID-4486

Signed-off-by: Giorgio Arena <giorgio.arena@arm.com>
Change-Id: Ib38b7943bd776a6d75d1da163908724c49eae73d
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/5864
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>

60 files changed, 187 insertions, 33 deletions

diff --git a/src/core/gpu/cl/kernels/ClActivationKernel.cpp b/src/core/gpu/cl/kernels/ClActivationKernel.cpp
index 17a8c6498d..e892d6a778 100644
--- a/src/core/gpu/cl/kernels/ClActivationKernel.cpp
+++ b/src/core/gpu/cl/kernels/ClActivationKernel.cpp
@@ -89,6 +89,7 @@ Status validate_arguments(const ITensorInfo *src, const ITensorInfo *dst, const
 ClActivationKernel::ClActivationKernel()
     : _run_in_place(false)
 {
+    _type = CLKernelType::ELEMENTWISE;
 }
 
 void ClActivationKernel::configure(const ClCompileContext &compile_context, ITensorInfo *src, ITensorInfo *dst, ActivationLayerInfo act_info)
diff --git a/src/core/gpu/cl/kernels/ClBatchConcatenateKernel.cpp b/src/core/gpu/cl/kernels/ClBatchConcatenateKernel.cpp
index 26f5113822..dbc628d6d5 100644
--- a/src/core/gpu/cl/kernels/ClBatchConcatenateKernel.cpp
+++ b/src/core/gpu/cl/kernels/ClBatchConcatenateKernel.cpp
@@ -62,6 +62,7 @@ Status validate_arguments(const ITensorInfo *src, unsigned int batch_offset, con
 ClBatchConcatenateKernel::ClBatchConcatenateKernel()
     : _batch_offset(0)
 {
+    _type = CLKernelType::ELEMENTWISE;
 }
 
 void ClBatchConcatenateKernel::configure(const CLCompileContext &compile_context, ITensorInfo *src, unsigned int batch_offset, ITensorInfo *dst)
diff --git a/src/core/gpu/cl/kernels/ClCastKernel.cpp b/src/core/gpu/cl/kernels/ClCastKernel.cpp
index 7a1d5c2824..fac9ebe5cf 100644
--- a/src/core/gpu/cl/kernels/ClCastKernel.cpp
+++ b/src/core/gpu/cl/kernels/ClCastKernel.cpp
@@ -72,6 +72,11 @@ Status validate_arguments(const ITensorInfo *src, const ITensorInfo *dst, Conver
 }
 } // namespace
 
+ClCastKernel::ClCastKernel()
+{
+    _type = CLKernelType::ELEMENTWISE;
+}
+
 void ClCastKernel::configure(const CLCompileContext &compile_context, const ITensorInfo *src, ITensorInfo *dst, ConvertPolicy policy)
 {
     ARM_COMPUTE_ERROR_ON_NULLPTR(src, dst);
diff --git a/src/core/gpu/cl/kernels/ClCastKernel.h b/src/core/gpu/cl/kernels/ClCastKernel.h
index 451aa9c1ab..6bf3cd9e50 100644
--- a/src/core/gpu/cl/kernels/ClCastKernel.h
+++ b/src/core/gpu/cl/kernels/ClCastKernel.h
@@ -41,7 +41,7 @@ namespace kernels
 class ClCastKernel : public IClKernel
 {
 public:
-    ClCastKernel() = default;
+    ClCastKernel();
     ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE(ClCastKernel);
     /** Set the src and dst of the kernel.
      *
diff --git a/src/core/gpu/cl/kernels/ClConvertFullyConnectedWeightsKernel.cpp b/src/core/gpu/cl/kernels/ClConvertFullyConnectedWeightsKernel.cpp
index 49f2f68a76..d1abd274d6 100644
--- a/src/core/gpu/cl/kernels/ClConvertFullyConnectedWeightsKernel.cpp
+++ b/src/core/gpu/cl/kernels/ClConvertFullyConnectedWeightsKernel.cpp
@@ -40,6 +40,11 @@ namespace opencl
 {
 namespace kernels
 {
+ClConvertFullyConnectedWeightsKernel::ClConvertFullyConnectedWeightsKernel()
+{
+    _type = CLKernelType::ELEMENTWISE;
+}
+
 void ClConvertFullyConnectedWeightsKernel::configure(const CLCompileContext &compile_context, const ITensorInfo *src, ITensorInfo *dst, const TensorShape &original_src_shape,
                                                      DataLayout data_layout)
 {
diff --git a/src/core/gpu/cl/kernels/ClConvertFullyConnectedWeightsKernel.h b/src/core/gpu/cl/kernels/ClConvertFullyConnectedWeightsKernel.h
index 11ab4d2a0d..6f4f09dc32 100644
--- a/src/core/gpu/cl/kernels/ClConvertFullyConnectedWeightsKernel.h
+++ b/src/core/gpu/cl/kernels/ClConvertFullyConnectedWeightsKernel.h
@@ -47,7 +47,7 @@ namespace kernels
 class ClConvertFullyConnectedWeightsKernel : public IClKernel
 {
 public:
-    ClConvertFullyConnectedWeightsKernel() = default;
+    ClConvertFullyConnectedWeightsKernel();
     ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE(ClConvertFullyConnectedWeightsKernel);
     /** Set the src and dst tensor.
      *
diff --git a/src/core/gpu/cl/kernels/ClCopyKernel.cpp b/src/core/gpu/cl/kernels/ClCopyKernel.cpp
index d6c87f8fad..98c6f34e60 100644
--- a/src/core/gpu/cl/kernels/ClCopyKernel.cpp
+++ b/src/core/gpu/cl/kernels/ClCopyKernel.cpp
@@ -68,6 +68,11 @@ Status validate_arguments(const ITensorInfo *src, const ITensorInfo *dst, Window
 
 } // namespace
 
+ClCopyKernel::ClCopyKernel()
+{
+    _type = CLKernelType::ELEMENTWISE;
+}
+
 void ClCopyKernel::configure(const CLCompileContext &compile_context, const ITensorInfo *src, ITensorInfo *dst, Window *dst_window)
 {
     ARM_COMPUTE_ERROR_ON_NULLPTR(src, dst);
diff --git a/src/core/gpu/cl/kernels/ClCopyKernel.h b/src/core/gpu/cl/kernels/ClCopyKernel.h
index b1b9672bcb..f3eb0aab62 100644
--- a/src/core/gpu/cl/kernels/ClCopyKernel.h
+++ b/src/core/gpu/cl/kernels/ClCopyKernel.h
@@ -38,7 +38,7 @@ namespace kernels
 class ClCopyKernel : public IClKernel
 {
 public:
-    ClCopyKernel() = default;
+    ClCopyKernel();
     ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE(ClCopyKernel);
     /** Initialize the kernel's src, dst.
      *
diff --git a/src/core/gpu/cl/kernels/ClCropKernel.cpp b/src/core/gpu/cl/kernels/ClCropKernel.cpp
index 1d322eefa1..ef2e48b45d 100644
--- a/src/core/gpu/cl/kernels/ClCropKernel.cpp
+++ b/src/core/gpu/cl/kernels/ClCropKernel.cpp
@@ -46,6 +46,11 @@ void ClCropKernel::configure(const ITensorInfo *src, ITensorInfo *dst, Coordinat
     configure(CLKernelLibrary::get().get_compile_context(), src, dst, start, end, batch_index, extrapolation_value, dst_window);
 }
 
+ClCropKernel::ClCropKernel()
+{
+    _type = CLKernelType::ELEMENTWISE;
+}
+
 void ClCropKernel::configure(const CLCompileContext &compile_context, const ITensorInfo *src, ITensorInfo *dst, Coordinates2D start, Coordinates2D end, uint32_t batch_index,
                              float extrapolation_value, Window *dst_window)
 {
diff --git a/src/core/gpu/cl/kernels/ClCropKernel.h b/src/core/gpu/cl/kernels/ClCropKernel.h
index ec0f8e58da..7120dbbc81 100644
--- a/src/core/gpu/cl/kernels/ClCropKernel.h
+++ b/src/core/gpu/cl/kernels/ClCropKernel.h
@@ -38,7 +38,7 @@ namespace kernels
 class ClCropKernel : public IClKernel
 {
 public:
-    ClCropKernel() = default;
+    ClCropKernel();
     ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE(ClCropKernel);
     /** Configure kernel
      *
diff --git a/src/core/gpu/cl/kernels/ClDepthConcatenateKernel.cpp b/src/core/gpu/cl/kernels/ClDepthConcatenateKernel.cpp
index 4039570da4..e3e384f748 100644
--- a/src/core/gpu/cl/kernels/ClDepthConcatenateKernel.cpp
+++ b/src/core/gpu/cl/kernels/ClDepthConcatenateKernel.cpp
@@ -61,6 +61,7 @@ Status validate_arguments(const ITensorInfo *src, unsigned int depth_offset, con
 ClDepthConcatenateKernel::ClDepthConcatenateKernel()
     : _depth_offset(0)
 {
+    _type = CLKernelType::ELEMENTWISE;
 }
 
 void ClDepthConcatenateKernel::configure(const CLCompileContext &compile_context, ITensorInfo *src, unsigned int depth_offset, ITensorInfo *dst)
diff --git a/src/core/gpu/cl/kernels/ClDequantizeKernel.cpp b/src/core/gpu/cl/kernels/ClDequantizeKernel.cpp
index f2758b759f..d69da8716c 100644
--- a/src/core/gpu/cl/kernels/ClDequantizeKernel.cpp
+++ b/src/core/gpu/cl/kernels/ClDequantizeKernel.cpp
@@ -61,6 +61,11 @@ Status validate_arguments(const ITensorInfo *src, const ITensorInfo *dst)
 }
 } // namespace
 
+ClDequantizeKernel::ClDequantizeKernel()
+{
+    _type = CLKernelType::ELEMENTWISE;
+}
+
 void ClDequantizeKernel::configure(const CLCompileContext &compile_context, ITensorInfo *src, ITensorInfo *dst)
 {
     ARM_COMPUTE_ERROR_ON_NULLPTR(src, dst);
diff --git a/src/core/gpu/cl/kernels/ClDequantizeKernel.h b/src/core/gpu/cl/kernels/ClDequantizeKernel.h
index 33e0164cc9..2460674067 100644
--- a/src/core/gpu/cl/kernels/ClDequantizeKernel.h
+++ b/src/core/gpu/cl/kernels/ClDequantizeKernel.h
@@ -39,7 +39,7 @@ class ClDequantizeKernel : public IClKernel
 {
 public:
     /** Default constructor */
-    ClDequantizeKernel() = default;
+    ClDequantizeKernel();
     ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE(ClDequantizeKernel);
     /** Initialise the kernel's input and output
      *
diff --git a/src/core/gpu/cl/kernels/ClDirectConv2dKernel.cpp b/src/core/gpu/cl/kernels/ClDirectConv2dKernel.cpp
index 94c4044bff..7b98671da2 100644
--- a/src/core/gpu/cl/kernels/ClDirectConv2dKernel.cpp
+++ b/src/core/gpu/cl/kernels/ClDirectConv2dKernel.cpp
@@ -377,6 +377,11 @@ BorderSize ClDirectConv2dKernel::border_size() const
     return _border_size;
 }
 
+ClDirectConv2dKernel::ClDirectConv2dKernel()
+{
+    _type = CLKernelType::DIRECT;
+}
+
 void ClDirectConv2dKernel::configure(const CLCompileContext &compile_context, ITensorInfo *src, ITensorInfo *weights, ITensorInfo *biases, ITensorInfo *dst,
                                      const PadStrideInfo &conv_info, const ActivationLayerInfo &act_info)
 {
diff --git a/src/core/gpu/cl/kernels/ClDirectConv2dKernel.h b/src/core/gpu/cl/kernels/ClDirectConv2dKernel.h
index e76666fd36..b592a2191b 100644
--- a/src/core/gpu/cl/kernels/ClDirectConv2dKernel.h
+++ b/src/core/gpu/cl/kernels/ClDirectConv2dKernel.h
@@ -38,7 +38,7 @@ namespace kernels
 class ClDirectConv2dKernel : public IClKernel
 {
 public:
-    ClDirectConv2dKernel() = default;
+    ClDirectConv2dKernel();
     ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE(ClDirectConv2dKernel);
     /** Set the src, weights, biases and dst tensors info.
      *
diff --git a/src/core/gpu/cl/kernels/ClElementwiseKernel.cpp b/src/core/gpu/cl/kernels/ClElementwiseKernel.cpp
index 335ee9c392..7bfdb9efdd 100644
--- a/src/core/gpu/cl/kernels/ClElementwiseKernel.cpp
+++ b/src/core/gpu/cl/kernels/ClElementwiseKernel.cpp
@@ -98,7 +98,7 @@ Status validate_arguments_with_float_only_supported_rules(const ITensorInfo &src
     return Status{};
 }
 
-Status validate_arguments_divide_operation(const ITensorInfo* src1, const ITensorInfo* src2, const ITensorInfo* dst)
+Status validate_arguments_divide_operation(const ITensorInfo *src1, const ITensorInfo *src2, const ITensorInfo *dst)
 {
     ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(src1, src2, dst);
     ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(src1);
@@ -271,6 +271,11 @@ std::pair<Status, Window> validate_and_configure_window_for_division(ITensorInfo
 }
 } // namespace
 
+ClElementwiseKernel::ClElementwiseKernel()
+{
+    _type = CLKernelType::ELEMENTWISE;
+}
+
 void ClElementwiseKernel::configure_common(ITensorInfo *src1, ITensorInfo *src2, ITensorInfo *dst)
 {
     configure_common(CLKernelLibrary::get().get_compile_context(), src1, src2, dst);
diff --git a/src/core/gpu/cl/kernels/ClElementwiseKernel.h b/src/core/gpu/cl/kernels/ClElementwiseKernel.h
index 4ed8ae73ab..7f55151a87 100644
--- a/src/core/gpu/cl/kernels/ClElementwiseKernel.h
+++ b/src/core/gpu/cl/kernels/ClElementwiseKernel.h
@@ -45,7 +45,7 @@ class ClElementwiseKernel : public IClKernel
 {
 public:
     /** Default constructor */
-    ClElementwiseKernel() = default;
+    ClElementwiseKernel();
     ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE(ClElementwiseKernel);
 
     // Inherited methods overridden:
@@ -89,9 +89,15 @@ protected:
     ActivationLayerInfo _act_info{};
 
 private:
-    const ITensorInfo *_src1{ nullptr }; /**< Source tensor info 1 */
-    const ITensorInfo *_src2{ nullptr }; /**< Source tensor info 2 */
-    ITensorInfo       *_dst{ nullptr };  /**< Destination tensor info */
+    const ITensorInfo *_src1
+    {
+        nullptr
+    }; /**< Source tensor info 1 */
+    const ITensorInfo *_src2
+    {
+        nullptr
+    };                            /**< Source tensor info 2 */
+    ITensorInfo *_dst{ nullptr }; /**< Destination tensor info */
 };
 
 class ClLogicalBinaryKernel : public ClElementwiseKernel
diff --git a/src/core/gpu/cl/kernels/ClElementwiseUnaryKernel.cpp b/src/core/gpu/cl/kernels/ClElementwiseUnaryKernel.cpp
index 5cbb3f2e38..1525c0fe54 100644
--- a/src/core/gpu/cl/kernels/ClElementwiseUnaryKernel.cpp
+++ b/src/core/gpu/cl/kernels/ClElementwiseUnaryKernel.cpp
@@ -66,6 +66,11 @@ Status validate_arguments(const ITensorInfo &src, const ITensorInfo &dst, const
 }
 } // namespace
 
+ClElementWiseUnaryKernel::ClElementWiseUnaryKernel()
+{
+    _type = CLKernelType::ELEMENTWISE;
+}
+
 void ClElementWiseUnaryKernel::configure(const CLCompileContext &compile_context, const ITensorInfo *src, ITensorInfo *dst, const ElementWiseUnary &op)
 {
     ARM_COMPUTE_ERROR_ON_NULLPTR(src, dst);
diff --git a/src/core/gpu/cl/kernels/ClElementwiseUnaryKernel.h b/src/core/gpu/cl/kernels/ClElementwiseUnaryKernel.h
index 7e5edef3ee..225869b58b 100644
--- a/src/core/gpu/cl/kernels/ClElementwiseUnaryKernel.h
+++ b/src/core/gpu/cl/kernels/ClElementwiseUnaryKernel.h
@@ -38,7 +38,7 @@ namespace kernels
 class ClElementWiseUnaryKernel : public IClKernel
 {
 public:
-    ClElementWiseUnaryKernel() = default;
+    ClElementWiseUnaryKernel();
     ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE(ClElementWiseUnaryKernel);
     /** Initialise the kernel's srcs, dst.
      *
diff --git a/src/core/gpu/cl/kernels/ClFillKernel.cpp b/src/core/gpu/cl/kernels/ClFillKernel.cpp
index b194ee549b..526a466a00 100644
--- a/src/core/gpu/cl/kernels/ClFillKernel.cpp
+++ b/src/core/gpu/cl/kernels/ClFillKernel.cpp
@@ -42,6 +42,11 @@ namespace opencl
 {
 namespace kernels
 {
+ClFillKernel::ClFillKernel()
+{
+    _type = CLKernelType::ELEMENTWISE;
+}
+
 void ClFillKernel::configure(ITensorInfo      *tensor,
                              const PixelValue &constant_value,
                              Window           *window)
diff --git a/src/core/gpu/cl/kernels/ClFillKernel.h b/src/core/gpu/cl/kernels/ClFillKernel.h
index b439eac0de..9542c20508 100644
--- a/src/core/gpu/cl/kernels/ClFillKernel.h
+++ b/src/core/gpu/cl/kernels/ClFillKernel.h
@@ -38,7 +38,7 @@ namespace kernels
 class ClFillKernel : public IClKernel
 {
 public:
-    ClFillKernel() = default;
+    ClFillKernel();
     ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE(ClFillKernel);
     /** Initialise the kernel's tensor and filling value
      *
diff --git a/src/core/gpu/cl/kernels/ClFloorKernel.cpp b/src/core/gpu/cl/kernels/ClFloorKernel.cpp
index 7296d40eaf..2047128963 100644
--- a/src/core/gpu/cl/kernels/ClFloorKernel.cpp
+++ b/src/core/gpu/cl/kernels/ClFloorKernel.cpp
@@ -61,6 +61,11 @@ Status validate_arguments(const ITensorInfo *src, const ITensorInfo *dst)
 }
 } // namespace
 
+ClFloorKernel::ClFloorKernel()
+{
+    _type = CLKernelType::ELEMENTWISE;
+}
+
 void ClFloorKernel::configure(const ClCompileContext &compile_context, const ITensorInfo *src, ITensorInfo *dst)
 {
     ARM_COMPUTE_ERROR_ON_NULLPTR(src, dst);
diff --git a/src/core/gpu/cl/kernels/ClFloorKernel.h b/src/core/gpu/cl/kernels/ClFloorKernel.h
index 646dfb30d8..3bc648b7be 100644
--- a/src/core/gpu/cl/kernels/ClFloorKernel.h
+++ b/src/core/gpu/cl/kernels/ClFloorKernel.h
@@ -38,7 +38,7 @@ namespace kernels
 class ClFloorKernel : public IClKernel
 {
 public:
-    ClFloorKernel() = default;
+    ClFloorKernel();
     ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE(ClFloorKernel);
     /** Configure kernel for a given list of arguments
      *
diff --git a/src/core/gpu/cl/kernels/ClGemmMatrixMultiplyKernel.cpp b/src/core/gpu/cl/kernels/ClGemmMatrixMultiplyKernel.cpp
index 817a105b14..6079644935 100644
--- a/src/core/gpu/cl/kernels/ClGemmMatrixMultiplyKernel.cpp
+++ b/src/core/gpu/cl/kernels/ClGemmMatrixMultiplyKernel.cpp
@@ -262,6 +262,11 @@ inline std::pair<Status, Window> validate_and_configure_window(ITensorInfo *src0
 }
 } // namespace
 
+ClGemmMatrixMultiplyKernel::ClGemmMatrixMultiplyKernel()
+{
+    _type = CLKernelType::GEMM;
+}
+
 void ClGemmMatrixMultiplyKernel::configure(const CLCompileContext &compile_context, ITensorInfo *src0, ITensorInfo *src1, ITensorInfo *src2, ITensorInfo *dst, float alpha,
                                            float beta,
                                            bool is_interleaved_transposed, const GEMMReshapeInfo &reshape_info, bool fp_mixed_precision, const ActivationLayerInfo &activation_info)
diff --git a/src/core/gpu/cl/kernels/ClGemmMatrixMultiplyKernel.h b/src/core/gpu/cl/kernels/ClGemmMatrixMultiplyKernel.h
index c1601335ee..c303f78b07 100644
--- a/src/core/gpu/cl/kernels/ClGemmMatrixMultiplyKernel.h
+++ b/src/core/gpu/cl/kernels/ClGemmMatrixMultiplyKernel.h
@@ -45,7 +45,7 @@ namespace kernels
 class ClGemmMatrixMultiplyKernel : public IClKernel
 {
 public:
-    ClGemmMatrixMultiplyKernel() = default;
+    ClGemmMatrixMultiplyKernel();
     ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE(ClGemmMatrixMultiplyKernel);
     /** Initialise the kernel's input, output and alpha
      *
diff --git a/src/core/gpu/cl/kernels/ClGemmMatrixMultiplyNativeKernel.cpp b/src/core/gpu/cl/kernels/ClGemmMatrixMultiplyNativeKernel.cpp
index 97d64c433c..5ae55ab04a 100644
--- a/src/core/gpu/cl/kernels/ClGemmMatrixMultiplyNativeKernel.cpp
+++ b/src/core/gpu/cl/kernels/ClGemmMatrixMultiplyNativeKernel.cpp
@@ -194,6 +194,11 @@ std::pair<Status, Window> validate_and_configure_window(ITensorInfo *src0, ITens
 }
 } // namespace
 
+ClGemmMatrixMultiplyNativeKernel::ClGemmMatrixMultiplyNativeKernel()
+{
+    _type = CLKernelType::GEMM;
+}
+
 void ClGemmMatrixMultiplyNativeKernel::configure(const CLCompileContext &compile_context, ITensorInfo *src0, ITensorInfo *src1, ITensorInfo *src2, ITensorInfo *dst, float alpha,
                                                  float                    beta,
                                                  const GEMMLHSMatrixInfo &lhs_info,
diff --git a/src/core/gpu/cl/kernels/ClGemmMatrixMultiplyNativeKernel.h b/src/core/gpu/cl/kernels/ClGemmMatrixMultiplyNativeKernel.h
index 4770b18b8e..c3bdc7589e 100644
--- a/src/core/gpu/cl/kernels/ClGemmMatrixMultiplyNativeKernel.h
+++ b/src/core/gpu/cl/kernels/ClGemmMatrixMultiplyNativeKernel.h
@@ -39,7 +39,7 @@ namespace kernels
 class ClGemmMatrixMultiplyNativeKernel : public IClKernel
 {
 public:
-    ClGemmMatrixMultiplyNativeKernel() = default;
+    ClGemmMatrixMultiplyNativeKernel();
     ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE(ClGemmMatrixMultiplyNativeKernel);
     /** Initialise the kernel's input and dst.
      *
diff --git a/src/core/gpu/cl/kernels/ClGemmMatrixMultiplyReshapedKernel.cpp b/src/core/gpu/cl/kernels/ClGemmMatrixMultiplyReshapedKernel.cpp
index 27409b66ac..591834f762 100644
--- a/src/core/gpu/cl/kernels/ClGemmMatrixMultiplyReshapedKernel.cpp
+++ b/src/core/gpu/cl/kernels/ClGemmMatrixMultiplyReshapedKernel.cpp
@@ -184,6 +184,11 @@ std::pair<Status, Window> validate_and_configure_window(ITensorInfo *src0, ITens
 }
 } // namespace
 
+ClGemmMatrixMultiplyReshapedKernel::ClGemmMatrixMultiplyReshapedKernel()
+{
+    _type = CLKernelType::GEMM;
+}
+
 void ClGemmMatrixMultiplyReshapedKernel::configure(const CLCompileContext &compile_context,
                                                    ITensorInfo *src0, ITensorInfo *src1, ITensorInfo *src2, ITensorInfo *dst, float alpha, float beta,
                                                    const GEMMLHSMatrixInfo &lhs_info, const GEMMRHSMatrixInfo &rhs_info, const GEMMKernelInfo &gemm_info)
diff --git a/src/core/gpu/cl/kernels/ClGemmMatrixMultiplyReshapedKernel.h b/src/core/gpu/cl/kernels/ClGemmMatrixMultiplyReshapedKernel.h
index ab648f15ae..b8ae4b9ae3 100644
--- a/src/core/gpu/cl/kernels/ClGemmMatrixMultiplyReshapedKernel.h
+++ b/src/core/gpu/cl/kernels/ClGemmMatrixMultiplyReshapedKernel.h
@@ -45,7 +45,7 @@ namespace kernels
 class ClGemmMatrixMultiplyReshapedKernel : public IClKernel
 {
 public:
-    ClGemmMatrixMultiplyReshapedKernel() = default;
+    ClGemmMatrixMultiplyReshapedKernel();
     ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE(ClGemmMatrixMultiplyReshapedKernel);
     /** Initialise the kernel's input and output.
      *
diff --git a/src/core/gpu/cl/kernels/ClGemmMatrixMultiplyReshapedOnlyRhsKernel.cpp b/src/core/gpu/cl/kernels/ClGemmMatrixMultiplyReshapedOnlyRhsKernel.cpp
index 4eea2c6f76..32ee0f9705 100644
--- a/src/core/gpu/cl/kernels/ClGemmMatrixMultiplyReshapedOnlyRhsKernel.cpp
+++ b/src/core/gpu/cl/kernels/ClGemmMatrixMultiplyReshapedOnlyRhsKernel.cpp
@@ -181,6 +181,11 @@ std::pair<Status, Window> validate_and_configure_window(ITensorInfo *src0, ITens
 }
 } // namespace
 
+ClGemmMatrixMultiplyReshapedOnlyRhsKernel::ClGemmMatrixMultiplyReshapedOnlyRhsKernel()
+{
+    _type = CLKernelType::GEMM;
+}
+
 void ClGemmMatrixMultiplyReshapedOnlyRhsKernel::configure(const CLCompileContext &compile_context,
                                                           ITensorInfo *src0, ITensorInfo *src1, ITensorInfo *src2, ITensorInfo *dst, float alpha, float beta,
                                                           const GEMMLHSMatrixInfo &lhs_info, const GEMMRHSMatrixInfo &rhs_info, const GEMMKernelInfo &gemm_info)
diff --git a/src/core/gpu/cl/kernels/ClGemmMatrixMultiplyReshapedOnlyRhsKernel.h b/src/core/gpu/cl/kernels/ClGemmMatrixMultiplyReshapedOnlyRhsKernel.h
index ff6c391e15..3d6164eca9 100644
--- a/src/core/gpu/cl/kernels/ClGemmMatrixMultiplyReshapedOnlyRhsKernel.h
+++ b/src/core/gpu/cl/kernels/ClGemmMatrixMultiplyReshapedOnlyRhsKernel.h
@@ -43,7 +43,7 @@ namespace kernels
 class ClGemmMatrixMultiplyReshapedOnlyRhsKernel : public ICLKernel
 {
 public:
-    ClGemmMatrixMultiplyReshapedOnlyRhsKernel() = default;
+    ClGemmMatrixMultiplyReshapedOnlyRhsKernel();
     ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE(ClGemmMatrixMultiplyReshapedOnlyRhsKernel);
     /** Initialise the kernel's input and output.
      *
diff --git a/src/core/gpu/cl/kernels/ClGemmReshapeLhsMatrixKernel.cpp b/src/core/gpu/cl/kernels/ClGemmReshapeLhsMatrixKernel.cpp
index 98161edfff..f92945e2a4 100644
--- a/src/core/gpu/cl/kernels/ClGemmReshapeLhsMatrixKernel.cpp
+++ b/src/core/gpu/cl/kernels/ClGemmReshapeLhsMatrixKernel.cpp
@@ -111,6 +111,11 @@ std::pair<Status, Window> validate_and_configure_window(ITensorInfo *src, ITenso
 }
 } // namespace
 
+ClGemmReshapeLhsMatrixKernel::ClGemmReshapeLhsMatrixKernel()
+{
+    _type = CLKernelType::ELEMENTWISE;
+}
+
 void ClGemmReshapeLhsMatrixKernel::configure(const CLCompileContext &compile_context, ITensorInfo *src, ITensorInfo *dst, const GEMMLHSMatrixInfo &lhs_info, bool reinterpret_input_as_3d)
 {
     ARM_COMPUTE_ERROR_ON_NULLPTR(src, dst);
diff --git a/src/core/gpu/cl/kernels/ClGemmReshapeLhsMatrixKernel.h b/src/core/gpu/cl/kernels/ClGemmReshapeLhsMatrixKernel.h
index b830ba02b4..73d811f3c3 100644
--- a/src/core/gpu/cl/kernels/ClGemmReshapeLhsMatrixKernel.h
+++ b/src/core/gpu/cl/kernels/ClGemmReshapeLhsMatrixKernel.h
@@ -41,7 +41,7 @@ namespace kernels
 class ClGemmReshapeLhsMatrixKernel : public ICLKernel
 {
 public:
-    ClGemmReshapeLhsMatrixKernel() = default;
+    ClGemmReshapeLhsMatrixKernel();
     ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE(ClGemmReshapeLhsMatrixKernel);
     /** Initialise the kernel's input and output.
      *
diff --git a/src/core/gpu/cl/kernels/ClGemmReshapeRhsMatrixKernel.cpp b/src/core/gpu/cl/kernels/ClGemmReshapeRhsMatrixKernel.cpp
index e1ef7c61aa..3a6f3c7e8f 100644
--- a/src/core/gpu/cl/kernels/ClGemmReshapeRhsMatrixKernel.cpp
+++ b/src/core/gpu/cl/kernels/ClGemmReshapeRhsMatrixKernel.cpp
@@ -107,6 +107,11 @@ std::pair<Status, Window> validate_and_configure_window(ITensorInfo *src, ITenso
 }
 } // namespace
 
+ClGemmReshapeRhsMatrixKernel::ClGemmReshapeRhsMatrixKernel()
+{
+    _type = CLKernelType::ELEMENTWISE;
+}
+
 void ClGemmReshapeRhsMatrixKernel::configure(const CLCompileContext &compile_context, ITensorInfo *src, ITensorInfo *dst, const GEMMRHSMatrixInfo &rhs_info)
 {
     ARM_COMPUTE_ERROR_ON_NULLPTR(src, dst);
diff --git a/src/core/gpu/cl/kernels/ClGemmReshapeRhsMatrixKernel.h b/src/core/gpu/cl/kernels/ClGemmReshapeRhsMatrixKernel.h
index e877d87408..27f80d3428 100644
--- a/src/core/gpu/cl/kernels/ClGemmReshapeRhsMatrixKernel.h
+++ b/src/core/gpu/cl/kernels/ClGemmReshapeRhsMatrixKernel.h
@@ -40,7 +40,7 @@ namespace kernels
 class ClGemmReshapeRhsMatrixKernel : public ICLKernel
 {
 public:
-    ClGemmReshapeRhsMatrixKernel() = default;
+    ClGemmReshapeRhsMatrixKernel();
     ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE(ClGemmReshapeRhsMatrixKernel);
     /** Initialise the kernel's input and output.
      *
diff --git a/src/core/gpu/cl/kernels/ClHeightConcatenateKernel.cpp b/src/core/gpu/cl/kernels/ClHeightConcatenateKernel.cpp
index 4436e98fe3..9ff30eedcd 100644
--- a/src/core/gpu/cl/kernels/ClHeightConcatenateKernel.cpp
+++ b/src/core/gpu/cl/kernels/ClHeightConcatenateKernel.cpp
@@ -64,6 +64,7 @@ Status validate_arguments(const ITensorInfo *src, unsigned int height_offset, co
 ClHeightConcatenateKernel::ClHeightConcatenateKernel()
     : _height_offset(0)
 {
+    _type = CLKernelType::ELEMENTWISE;
 }
 
 Status ClHeightConcatenateKernel::validate(const ITensorInfo *src, unsigned int height_offset, const ITensorInfo *dst)
diff --git a/src/core/gpu/cl/kernels/ClMulKernel.cpp b/src/core/gpu/cl/kernels/ClMulKernel.cpp
index b8081bbacf..65f3bec099 100644
--- a/src/core/gpu/cl/kernels/ClMulKernel.cpp
+++ b/src/core/gpu/cl/kernels/ClMulKernel.cpp
@@ -92,6 +92,11 @@ Status validate_arguments(const ITensorInfo *src1, const ITensorInfo *src2, cons
 }
 } // namespace
 
+ClMulKernel::ClMulKernel()
+{
+    _type = CLKernelType::ELEMENTWISE;
+}
+
 void ClMulKernel::configure(const CLCompileContext &compile_context, ITensorInfo *src1, ITensorInfo *src2, ITensorInfo *dst, float scale,
                             ConvertPolicy overflow_policy, RoundingPolicy rounding_policy, const ActivationLayerInfo &act_info)
 {
@@ -128,7 +133,7 @@ void ClMulKernel::configure(const CLCompileContext &compile_context, ITensorInfo
     else
     {
         if(src1->element_size() == 4 || src2->element_size() == 4)
-        {   
+        {
             // use 64 bit accumulator for 32-bit input
             acc_type = "long";
         }
@@ -316,6 +321,11 @@ Status validate_arguments_complex(const ITensorInfo *src1, const ITensorInfo *sr
 }
 } // namespace
 
+ClComplexMulKernel::ClComplexMulKernel()
+{
+    _type = CLKernelType::ELEMENTWISE;
+}
+
 void ClComplexMulKernel::configure(const CLCompileContext &compile_context, ITensorInfo *src1, ITensorInfo *src2, ITensorInfo *dst, const ActivationLayerInfo &act_info)
 {
     ARM_COMPUTE_ERROR_ON_NULLPTR(src1, src2, dst);
diff --git a/src/core/gpu/cl/kernels/ClMulKernel.h b/src/core/gpu/cl/kernels/ClMulKernel.h
index 44162f3db3..9c70301d89 100644
--- a/src/core/gpu/cl/kernels/ClMulKernel.h
+++ b/src/core/gpu/cl/kernels/ClMulKernel.h
@@ -39,7 +39,7 @@ class ClMulKernel : public IClKernel
 {
 public:
     /** Default constructor */
-    ClMulKernel() = default;
+    ClMulKernel();
     ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE(ClMulKernel);
     /** Initialise the kernel's src and dst.
      *
@@ -88,7 +88,7 @@ class ClComplexMulKernel : public ICLKernel
 {
 public:
     /** Default constructor */
-    ClComplexMulKernel() = default;
+    ClComplexMulKernel();
     ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE(ClComplexMulKernel);
     /** Initialise the kernel's src and dst.
      *
diff --git a/src/core/gpu/cl/kernels/ClPermuteKernel.cpp b/src/core/gpu/cl/kernels/ClPermuteKernel.cpp
index ffc13060a8..722bf454f2 100644
--- a/src/core/gpu/cl/kernels/ClPermuteKernel.cpp
+++ b/src/core/gpu/cl/kernels/ClPermuteKernel.cpp
@@ -77,6 +77,11 @@ Status validate_arguments(const ITensorInfo *src, const ITensorInfo *dst, const
 }
 } // namespace
 
+ClPermuteKernel::ClPermuteKernel()
+{
+    _type = CLKernelType::ELEMENTWISE;
+}
+
 void ClPermuteKernel::configure(const CLCompileContext &compile_context, const ITensorInfo *src, ITensorInfo *dst, const PermutationVector &perm)
 {
     ARM_COMPUTE_ERROR_ON_NULLPTR(src, dst);
diff --git a/src/core/gpu/cl/kernels/ClPermuteKernel.h b/src/core/gpu/cl/kernels/ClPermuteKernel.h
index b844214595..326110a27c 100644
--- a/src/core/gpu/cl/kernels/ClPermuteKernel.h
+++ b/src/core/gpu/cl/kernels/ClPermuteKernel.h
@@ -42,7 +42,7 @@ class ClPermuteKernel : public IClKernel
 {
 public:
     /** Default constructor */
-    ClPermuteKernel() = default;
+    ClPermuteKernel();
     ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE(ClPermuteKernel);
     /** Set the src and dst of the kernel.
      *
diff --git a/src/core/gpu/cl/kernels/ClPool2dKernel.cpp b/src/core/gpu/cl/kernels/ClPool2dKernel.cpp
index 0e15bffd14..9d5a24fdf2 100644
--- a/src/core/gpu/cl/kernels/ClPool2dKernel.cpp
+++ b/src/core/gpu/cl/kernels/ClPool2dKernel.cpp
@@ -205,6 +205,7 @@ std::tuple<Status, Window, ClPoolingConfig> validate_and_configure_window(ITenso
 ClPool2dKernel::ClPool2dKernel()
     : _pool_info(), _data_layout(DataLayout::UNKNOWN), _border_size(0), _num_elems_processed_per_iteration(1)
 {
+    _type = CLKernelType::POOL;
 }
 
 BorderSize ClPool2dKernel::border_size() const
diff --git a/src/core/gpu/cl/kernels/ClQuantizeKernel.cpp b/src/core/gpu/cl/kernels/ClQuantizeKernel.cpp
index 48d351d536..7900489db7 100644
--- a/src/core/gpu/cl/kernels/ClQuantizeKernel.cpp
+++ b/src/core/gpu/cl/kernels/ClQuantizeKernel.cpp
@@ -61,6 +61,11 @@ Status validate_arguments(const ITensorInfo *src, const ITensorInfo *dst)
 }
 } // namespace
 
+ClQuantizeKernel::ClQuantizeKernel()
+{
+    _type = CLKernelType::ELEMENTWISE;
+}
+
 void ClQuantizeKernel::configure(const CLCompileContext &compile_context, const ITensorInfo *src, ITensorInfo *dst)
 {
     ARM_COMPUTE_ERROR_ON_NULLPTR(src, dst);
diff --git a/src/core/gpu/cl/kernels/ClQuantizeKernel.h b/src/core/gpu/cl/kernels/ClQuantizeKernel.h
index 8d37f33032..cd972987f5 100644
--- a/src/core/gpu/cl/kernels/ClQuantizeKernel.h
+++ b/src/core/gpu/cl/kernels/ClQuantizeKernel.h
@@ -42,7 +42,7 @@ class ClQuantizeKernel : public IClKernel
 {
 public:
     /** Default constructor */
-    ClQuantizeKernel() = default;
+    ClQuantizeKernel();
     ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE(ClQuantizeKernel);
     /** Set the input, output.
      *
diff --git a/src/core/gpu/cl/kernels/ClReshapeKernel.cpp b/src/core/gpu/cl/kernels/ClReshapeKernel.cpp
index 923b9cb264..fcda061930 100644
--- a/src/core/gpu/cl/kernels/ClReshapeKernel.cpp
+++ b/src/core/gpu/cl/kernels/ClReshapeKernel.cpp
@@ -62,6 +62,11 @@ Status validate_arguments(const ITensorInfo *src, const ITensorInfo *dst)
 }
 } // namespace
 
+ClReshapeKernel::ClReshapeKernel()
+{
+    _type = CLKernelType::ELEMENTWISE;
+}
+
 void ClReshapeKernel::configure(const CLCompileContext &compile_context, const ITensorInfo *src, ITensorInfo *dst)
 {
     ARM_COMPUTE_ERROR_ON_NULLPTR(src, dst);
diff --git a/src/core/gpu/cl/kernels/ClReshapeKernel.h b/src/core/gpu/cl/kernels/ClReshapeKernel.h
index 0501b93f40..3cd8369012 100644
--- a/src/core/gpu/cl/kernels/ClReshapeKernel.h
+++ b/src/core/gpu/cl/kernels/ClReshapeKernel.h
@@ -38,7 +38,7 @@ namespace kernels
 class ClReshapeKernel : public IClKernel
 {
 public:
-    ClReshapeKernel() = default;
+    ClReshapeKernel();
     ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE(ClReshapeKernel);
     /** Set the src and dst of the kernel
      *
diff --git a/src/core/gpu/cl/kernels/ClScaleKernel.cpp b/src/core/gpu/cl/kernels/ClScaleKernel.cpp
index 7fb5d2a5d3..57ca331539 100644
--- a/src/core/gpu/cl/kernels/ClScaleKernel.cpp
+++ b/src/core/gpu/cl/kernels/ClScaleKernel.cpp
@@ -140,6 +140,11 @@ Status ClScaleKernel::validate(const ITensorInfo *src, const ITensorInfo *dst, c
     return Status{};
 }
 
+ClScaleKernel::ClScaleKernel()
+{
+    _type = CLKernelType::ELEMENTWISE;
+}
+
 void ClScaleKernel::configure(const CLCompileContext &compile_context, ITensorInfo *src, ITensorInfo *dst, const ScaleKernelInfo &info)
 {
     ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(src, dst, info));
diff --git a/src/core/gpu/cl/kernels/ClScaleKernel.h b/src/core/gpu/cl/kernels/ClScaleKernel.h
index ad7632c713..826c4821b5 100644
--- a/src/core/gpu/cl/kernels/ClScaleKernel.h
+++ b/src/core/gpu/cl/kernels/ClScaleKernel.h
@@ -42,7 +42,7 @@ class ClScaleKernel : public IClKernel
 {
 public:
     /** Default constructor */
-    ClScaleKernel() = default;
+    ClScaleKernel();
     ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE(ClScaleKernel);
 
     /** Initialise the kernel's inputs, output and interpolation policy
diff --git a/src/core/gpu/cl/kernels/ClSoftmaxKernel.cpp b/src/core/gpu/cl/kernels/ClSoftmaxKernel.cpp
index 000c9ad04d..1dd905d66e 100644
--- a/src/core/gpu/cl/kernels/ClSoftmaxKernel.cpp
+++ b/src/core/gpu/cl/kernels/ClSoftmaxKernel.cpp
@@ -154,6 +154,11 @@ const unsigned int ClLogits1DMaxShiftExpSumKernel::_serial_vector_size = 8;
 /**< Vector size in the parallel case (obtained through auto-tuning, enables the best memory access pattern for Bifrost) .*/
 const unsigned int ClLogits1DMaxShiftExpSumKernel::_parallel_vector_size = 4;
 
+ClLogits1DMaxShiftExpSumKernel::ClLogits1DMaxShiftExpSumKernel()
+{
+    _type = CLKernelType::ELEMENTWISE;
+}
+
 void ClLogits1DMaxShiftExpSumKernel::configure(const CLCompileContext &compile_context, const ITensorInfo &src, ITensorInfo &max, ITensorInfo &dst, ITensorInfo &sum, const SoftmaxKernelInfo &info)
 {
     auto padding_info = get_padding_info({ &src, &max, &dst, &sum });
@@ -273,6 +278,11 @@ void ClLogits1DMaxShiftExpSumKernel::run_op(ITensorPack &tensors, const Window &
     while(window_collapsed.slide_window_slice_3D(slice));
 }
 
+ClLogits1DNormKernel::ClLogits1DNormKernel()
+{
+    _type = CLKernelType::ELEMENTWISE;
+}
+
 void ClLogits1DNormKernel::configure(const CLCompileContext &compile_context, const ITensorInfo &src, const ITensorInfo &sum, ITensorInfo &dst, const SoftmaxKernelInfo &info)
 {
     auto padding_info = get_padding_info({ &src, &dst, &sum });
diff --git a/src/core/gpu/cl/kernels/ClSoftmaxKernel.h b/src/core/gpu/cl/kernels/ClSoftmaxKernel.h
index af980eaa8e..db1aca3b9b 100644
--- a/src/core/gpu/cl/kernels/ClSoftmaxKernel.h
+++ b/src/core/gpu/cl/kernels/ClSoftmaxKernel.h
@@ -51,7 +51,7 @@ public:
     using ParallelReductionInfo = std::tuple<bool, unsigned int>;
 
     /** Default constructor */
-    ClLogits1DMaxShiftExpSumKernel() = default;
+    ClLogits1DMaxShiftExpSumKernel();
     ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE(ClLogits1DMaxShiftExpSumKernel);
     /** Configure the kernel using the given information about tensors
      *
@@ -94,7 +94,7 @@ class ClLogits1DNormKernel : public IClKernel
 {
 public:
     /** Default constructor */
-    ClLogits1DNormKernel() = default;
+    ClLogits1DNormKernel();
     ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE(ClLogits1DNormKernel);
 
     /** Set the input and output tensors.
diff --git a/src/core/gpu/cl/kernels/ClTransposeKernel.cpp b/src/core/gpu/cl/kernels/ClTransposeKernel.cpp
index 704d0152cf..40bd4b034a 100644
--- a/src/core/gpu/cl/kernels/ClTransposeKernel.cpp
+++ b/src/core/gpu/cl/kernels/ClTransposeKernel.cpp
@@ -43,6 +43,11 @@ namespace opencl
 {
 namespace kernels
 {
+ClTransposeKernel::ClTransposeKernel()
+{
+    _type = CLKernelType::ELEMENTWISE;
+}
+
 void ClTransposeKernel::configure(const CLCompileContext &compile_context, const ITensorInfo *src, ITensorInfo *dst)
 {
     ARM_COMPUTE_ERROR_ON_NULLPTR(src, dst);
diff --git a/src/core/gpu/cl/kernels/ClTransposeKernel.h b/src/core/gpu/cl/kernels/ClTransposeKernel.h
index 21d4fd41f5..7d1226cc0d 100644
--- a/src/core/gpu/cl/kernels/ClTransposeKernel.h
+++ b/src/core/gpu/cl/kernels/ClTransposeKernel.h
@@ -38,7 +38,7 @@ namespace kernels
 class ClTransposeKernel : public IClKernel
 {
 public:
-    ClTransposeKernel() = default;
+    ClTransposeKernel();
     ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE(ClTransposeKernel);
     /** Set the src and dst of the kernel.
      *
diff --git a/src/core/gpu/cl/kernels/ClWidthConcatenate2TensorsKernel.cpp b/src/core/gpu/cl/kernels/ClWidthConcatenate2TensorsKernel.cpp
index 9f970719ed..8607620e92 100644
--- a/src/core/gpu/cl/kernels/ClWidthConcatenate2TensorsKernel.cpp
+++ b/src/core/gpu/cl/kernels/ClWidthConcatenate2TensorsKernel.cpp
@@ -68,6 +68,11 @@ Status ClWidthConcatenate2TensorsKernel::validate(const ITensorInfo *src1, const
     return Status{};
 }
 
+ClWidthConcatenate2TensorsKernel::ClWidthConcatenate2TensorsKernel()
+{
+    _type = CLKernelType::ELEMENTWISE;
+}
+
 void ClWidthConcatenate2TensorsKernel::configure(const CLCompileContext &compile_context, ITensorInfo *src1, ITensorInfo *src2, ITensorInfo *dst)
 {
     ARM_COMPUTE_ERROR_ON_NULLPTR(src1, src2, dst);
diff --git a/src/core/gpu/cl/kernels/ClWidthConcatenate2TensorsKernel.h b/src/core/gpu/cl/kernels/ClWidthConcatenate2TensorsKernel.h
index ddade29113..56202bad2e 100644
--- a/src/core/gpu/cl/kernels/ClWidthConcatenate2TensorsKernel.h
+++ b/src/core/gpu/cl/kernels/ClWidthConcatenate2TensorsKernel.h
@@ -41,7 +41,7 @@ class ClWidthConcatenate2TensorsKernel : public IClKernel
 {
 public:
     /** Default constructor */
-    ClWidthConcatenate2TensorsKernel() = default;
+    ClWidthConcatenate2TensorsKernel();
     ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE(ClWidthConcatenate2TensorsKernel);
     /** Initialise the kernel's sources and destination
      *
diff --git a/src/core/gpu/cl/kernels/ClWidthConcatenate4TensorsKernel.cpp b/src/core/gpu/cl/kernels/ClWidthConcatenate4TensorsKernel.cpp
index 281d190381..edbc23c1d3 100644
--- a/src/core/gpu/cl/kernels/ClWidthConcatenate4TensorsKernel.cpp
+++ b/src/core/gpu/cl/kernels/ClWidthConcatenate4TensorsKernel.cpp
@@ -66,6 +66,7 @@ Status validate_arguments(const ITensorInfo *src1, const ITensorInfo *src2, cons
 
 ClWidthConcatenate4TensorsKernel::ClWidthConcatenate4TensorsKernel()
 {
+    _type = CLKernelType::ELEMENTWISE;
 }
 
 Status ClWidthConcatenate4TensorsKernel::validate(const ITensorInfo *src1, const ITensorInfo *src2, const ITensorInfo *src3, const ITensorInfo *src4, const ITensorInfo *dst)
diff --git a/src/core/gpu/cl/kernels/ClWidthConcatenateKernel.cpp b/src/core/gpu/cl/kernels/ClWidthConcatenateKernel.cpp
index d188a5226b..5510c746f8 100644
--- a/src/core/gpu/cl/kernels/ClWidthConcatenateKernel.cpp
+++ b/src/core/gpu/cl/kernels/ClWidthConcatenateKernel.cpp
@@ -63,6 +63,7 @@ Status validate_arguments(const ITensorInfo *src, unsigned int width_offset, con
 
 ClWidthConcatenateKernel::ClWidthConcatenateKernel()
 {
+    _type = CLKernelType::ELEMENTWISE;
 }
 
 Status ClWidthConcatenateKernel::validate(const ITensorInfo *src, unsigned int width_offset, const ITensorInfo *dst)
diff --git a/src/core/gpu/cl/kernels/ClWinogradFilterTransformKernel.cpp b/src/core/gpu/cl/kernels/ClWinogradFilterTransformKernel.cpp
index 381b4bcae9..ae43fed12d 100644
--- a/src/core/gpu/cl/kernels/ClWinogradFilterTransformKernel.cpp
+++ b/src/core/gpu/cl/kernels/ClWinogradFilterTransformKernel.cpp
@@ -91,6 +91,11 @@ std::pair<Status, Window> validate_and_configure_window(ITensorInfo *input, ITen
 }
 } // namespace
 
+ClWinogradFilterTransformKernel::ClWinogradFilterTransformKernel()
+{
+    _type = CLKernelType::WINOGRAD;
+}
+
 void ClWinogradFilterTransformKernel::configure(const ClCompileContext &compile_context, ITensorInfo *src, ITensorInfo *dst, const WinogradInfo &winograd_info)
 {
     ARM_COMPUTE_ERROR_ON_NULLPTR(src, dst);
diff --git a/src/core/gpu/cl/kernels/ClWinogradFilterTransformKernel.h b/src/core/gpu/cl/kernels/ClWinogradFilterTransformKernel.h
index 2bc2ceb36e..13200dc419 100644
--- a/src/core/gpu/cl/kernels/ClWinogradFilterTransformKernel.h
+++ b/src/core/gpu/cl/kernels/ClWinogradFilterTransformKernel.h
@@ -40,7 +40,7 @@ class ClWinogradFilterTransformKernel : public IClKernel
 {
 public:
     /** Default constructor */
-    ClWinogradFilterTransformKernel() = default;
+    ClWinogradFilterTransformKernel();
     ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE(ClWinogradFilterTransformKernel);
     /** Set the input and output tensor.
      *
diff --git a/src/core/gpu/cl/kernels/ClWinogradInputTransformKernel.cpp b/src/core/gpu/cl/kernels/ClWinogradInputTransformKernel.cpp
index 17f0eb9e2c..62db2282e0 100644
--- a/src/core/gpu/cl/kernels/ClWinogradInputTransformKernel.cpp
+++ b/src/core/gpu/cl/kernels/ClWinogradInputTransformKernel.cpp
@@ -103,6 +103,7 @@ std::pair<Status, Window> validate_and_configure_window(ITensorInfo *input, ITen
 ClWinogradInputTransformKernel::ClWinogradInputTransformKernel()
     : _border_size(0), _data_layout(DataLayout::UNKNOWN), _num_tiles_x(0), _num_tiles_y(0), _step_z(1)
 {
+    _type = CLKernelType::WINOGRAD;
 }
 
 BorderSize ClWinogradInputTransformKernel::border_size() const
diff --git a/src/core/gpu/cl/kernels/ClWinogradOutputTransformKernel.cpp b/src/core/gpu/cl/kernels/ClWinogradOutputTransformKernel.cpp
index a6c05420ed..f6ade57e5d 100644
--- a/src/core/gpu/cl/kernels/ClWinogradOutputTransformKernel.cpp
+++ b/src/core/gpu/cl/kernels/ClWinogradOutputTransformKernel.cpp
@@ -122,6 +122,11 @@ std::pair<Status, Window> validate_and_configure_window(ITensorInfo *input, ITen
 }
 } // namespace
 
+ClWinogradOutputTransformKernel::ClWinogradOutputTransformKernel()
+{
+    _type = CLKernelType::WINOGRAD;
+}
+
 void ClWinogradOutputTransformKernel::configure(const ClCompileContext &compile_context, ITensorInfo *src, ITensorInfo *bias, ITensorInfo *dst, const WinogradInfo &winograd_info,
                                                 const ActivationLayerInfo &act_info)
 {
diff --git a/src/core/gpu/cl/kernels/ClWinogradOutputTransformKernel.h b/src/core/gpu/cl/kernels/ClWinogradOutputTransformKernel.h
index 48b27e658c..2948d3f181 100644
--- a/src/core/gpu/cl/kernels/ClWinogradOutputTransformKernel.h
+++ b/src/core/gpu/cl/kernels/ClWinogradOutputTransformKernel.h
@@ -40,7 +40,7 @@ class ClWinogradOutputTransformKernel : public IClKernel
 {
 public:
     /** Default constructor */
-    ClWinogradOutputTransformKernel() = default;
+    ClWinogradOutputTransformKernel();
     ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE(ClWinogradOutputTransformKernel);
     /** Set the input and output tensor.
      *