IVGCVSW-1988: Refactor backend-specific unit tests

Change-Id: I8eca81d2e0780390eaa837c186ffe1c7d41fdebe

11 files changed, 6 insertions, 3061 deletions

diff --git a/src/backends/test/ActivationFixture.hpp b/src/backends/test/ActivationFixture.hpp
index d9d4ca7470..5028b252e1 100644
--- a/src/backends/test/ActivationFixture.hpp
+++ b/src/backends/test/ActivationFixture.hpp
@@ -7,6 +7,11 @@
 #include "TensorCopyUtils.hpp"
 #include "WorkloadTestUtils.hpp"
 
+#include <armnn/test/TensorHelpers.hpp>
+
+#include <boost/numeric/conversion/cast.hpp>
+#include <boost/multi_array.hpp>
+
 struct ActivationFixture
 {
     ActivationFixture()
diff --git a/src/backends/test/ArmComputeCl.cpp b/src/backends/test/ArmComputeCl.cpp
deleted file mode 100644
index f4ead34c58..0000000000
--- a/src/backends/test/ArmComputeCl.cpp
+++ /dev/null
@@ -1,335 +0,0 @@
-//
-// Copyright © 2017 Arm Ltd. All rights reserved.
-// SPDX-License-Identifier: MIT
-//
-#include <boost/test/unit_test.hpp>
-#include "test/TensorHelpers.hpp"
-#include "LayerTests.hpp"
-
-#include <backends/CpuTensorHandle.hpp>
-#include <backends/cl/ClWorkloadFactory.hpp>
-#include <backends/cl/workloads/ClWorkloadUtils.hpp>
-#include <backends/reference/RefWorkloadFactory.hpp>
-#include <backends/cl/ClLayerSupport.hpp>
-#include "ActivationFixture.hpp"
-#include "ClContextControlFixture.hpp"
-
-#include <arm_compute/core/CL/CLKernelLibrary.h>
-#include <arm_compute/runtime/CL/CLScheduler.h>
-#include <string>
-#include <iostream>
-
-#include "test/UnitTests.hpp"
-
-BOOST_FIXTURE_TEST_SUITE(Compute_ArmComputeCl, ClContextControlFixture)
-using FactoryType = armnn::ClWorkloadFactory;
-
-// ============================================================================
-// UNIT tests
-
-// Activation
-ARMNN_AUTO_TEST_CASE(ConstantLinearActivation, ConstantLinearActivationTest)
-
-ARMNN_AUTO_TEST_CASE(SimpleSoftmaxBeta1, SimpleSoftmaxTest, 1.0f)
-ARMNN_AUTO_TEST_CASE(SimpleSoftmaxBeta2, SimpleSoftmaxTest, 2.0f)
-ARMNN_AUTO_TEST_CASE(SimpleSoftmaxBeta1Uint8, SimpleSoftmaxUint8Test, 1.0f)
-ARMNN_AUTO_TEST_CASE(SimpleSoftmaxBeta2Uint8, SimpleSoftmaxUint8Test, 2.0f)
-
-ARMNN_AUTO_TEST_CASE(ReLu1Uint8, BoundedReLuUint8UpperAndLowerBoundTest)
-ARMNN_AUTO_TEST_CASE(ReLu6Uint8, BoundedReLuUint8UpperBoundOnlyTest)
-
-// Fully Connected
-ARMNN_AUTO_TEST_CASE(SimpleFullyConnected, FullyConnectedFloat32Test, false, false)
-ARMNN_AUTO_TEST_CASE(SimpleFullyConnectedWithBias, FullyConnectedFloat32Test, true, false)
-ARMNN_AUTO_TEST_CASE(SimpleFullyConnectedWithTranspose, FullyConnectedFloat32Test, false, true)
-ARMNN_AUTO_TEST_CASE(FullyConnectedUint8, FullyConnectedUint8Test, false)
-ARMNN_AUTO_TEST_CASE(FullyConnectedBiasedUint8, FullyConnectedUint8Test, true)
-
-ARMNN_AUTO_TEST_CASE(FullyConnectedLarge, FullyConnectedLargeTest, false)
-ARMNN_AUTO_TEST_CASE(FullyConnectedLargeTransposed, FullyConnectedLargeTest, true)
-
-// Convolution
-ARMNN_AUTO_TEST_CASE(SimpleConvolution1d, Convolution1dTest, true)
-
-ARMNN_AUTO_TEST_CASE(SimpleConvolution2d, SimpleConvolution2d3x5Test, true)
-ARMNN_AUTO_TEST_CASE(SimpleConvolution2dSquare, SimpleConvolution2d3x3Test, true)
-ARMNN_AUTO_TEST_CASE(SimpleConvolution2d3x3Uint8, SimpleConvolution2d3x3Uint8Test, true)
-ARMNN_AUTO_TEST_CASE(UnbiasedConvolution2d, SimpleConvolution2d3x5Test, false)
-ARMNN_AUTO_TEST_CASE(UnbiasedConvolution2dSquare, SimpleConvolution2d3x3Test, false)
-ARMNN_AUTO_TEST_CASE(SimpleConvolution2dAsymmetricPadding, Convolution2dAsymmetricPaddingTest)
-
-ARMNN_AUTO_TEST_CASE(SimpleConvolution2dSquareNhwc, SimpleConvolution2d3x3NhwcTest, false)
-
-// Depthwise Convolution
-ARMNN_AUTO_TEST_CASE(DepthwiseConvolution2dDepthMul1, DepthwiseConvolution2dDepthMul1Test, true)
-ARMNN_AUTO_TEST_CASE(UnbiasedDepthwiseConvolution2dDepthMul1, DepthwiseConvolution2dDepthMul1Test, false)
-ARMNN_AUTO_TEST_CASE(DepthwiseConvolution2dDepthMul1Uint8, DepthwiseConvolution2dDepthMul1Uint8Test, true)
-ARMNN_AUTO_TEST_CASE(UnbiasedDepthwiseConvolution2dDepthMul1Uint8, DepthwiseConvolution2dDepthMul1Uint8Test, false)
-
-ARMNN_AUTO_TEST_CASE(DepthwiseConvolution2dAsymmetric, DepthwiseConvolution2dAsymmetricTest, true)
-ARMNN_AUTO_TEST_CASE(UnbiasedDepthwiseConvolution2dAsymmetric, DepthwiseConvolution2dAsymmetricTest, false)
-
-// Softmax
-BOOST_AUTO_TEST_CASE(Softmax4dSupport)
-{
-    const unsigned int numDimensions = 4u;
-    std::array<unsigned int, numDimensions> dimensionSizes;
-    dimensionSizes.fill(1u);
-
-    const armnn::TensorInfo inputInfo(numDimensions, &dimensionSizes.front(), armnn::DataType::Float32);
-    const armnn::TensorInfo outputInfo(numDimensions, &dimensionSizes.front(), armnn::DataType::Float32);
-
-    // 4D Softmax should be reported as unsupported on the CL backend
-    BOOST_TEST(!armnn::IsSoftmaxSupportedCl(inputInfo, outputInfo, armnn::SoftmaxDescriptor()));
-}
-
-// Splitter
-ARMNN_AUTO_TEST_CASE(SimpleSplitter, SplitterTest)
-ARMNN_AUTO_TEST_CASE(SimpleSplitterUint8, SplitterUint8Test)
-
-ARMNN_AUTO_TEST_CASE(CopyViaSplitter, CopyViaSplitterTest)
-ARMNN_AUTO_TEST_CASE(CopyViaSplitterUint8, CopyViaSplitterUint8Test)
-
-// Merger
-ARMNN_AUTO_TEST_CASE(SimpleMerger, MergerTest)
-ARMNN_AUTO_TEST_CASE(MergerUint8, MergerUint8Test)
-
-// Normalization
-ARMNN_AUTO_TEST_CASE(SimpleNormalizationAcross, SimpleNormalizationAcrossTest)
-ARMNN_AUTO_TEST_CASE(SimpleNormalizationWithin, SimpleNormalizationWithinTest)
-ARMNN_AUTO_TEST_CASE(SimpleNormalizationAcrossNhwc, SimpleNormalizationAcrossNhwcTest)
-
-// Pooling
-ARMNN_AUTO_TEST_CASE(SimpleMaxPooling2dSize3x3Stride2x4, SimpleMaxPooling2dSize3x3Stride2x4Test, true)
-ARMNN_AUTO_TEST_CASE(SimpleMaxPooling2dSize3x3Stride2x4Uint8, SimpleMaxPooling2dSize3x3Stride2x4Uint8Test, true)
-
-ARMNN_AUTO_TEST_CASE(IgnorePaddingSimpleMaxPooling2d, IgnorePaddingSimpleMaxPooling2dTest)
-ARMNN_AUTO_TEST_CASE(IgnorePaddingSimpleMaxPooling2dUint8, IgnorePaddingSimpleMaxPooling2dUint8Test)
-ARMNN_AUTO_TEST_CASE(IgnorePaddingMaxPooling2dSize3, IgnorePaddingMaxPooling2dSize3Test)
-ARMNN_AUTO_TEST_CASE(IgnorePaddingMaxPooling2dSize3Uint8, IgnorePaddingMaxPooling2dSize3Uint8Test)
-
-ARMNN_AUTO_TEST_CASE(IgnorePaddingSimpleAveragePooling2d, IgnorePaddingSimpleAveragePooling2dTest)
-ARMNN_AUTO_TEST_CASE(IgnorePaddingSimpleAveragePooling2dUint8, IgnorePaddingSimpleAveragePooling2dUint8Test)
-ARMNN_AUTO_TEST_CASE(IgnorePaddingSimpleAveragePooling2dNoPadding, IgnorePaddingSimpleAveragePooling2dNoPaddingTest)
-ARMNN_AUTO_TEST_CASE(IgnorePaddingSimpleAveragePooling2dNoPaddingUint8,
-                     IgnorePaddingSimpleAveragePooling2dNoPaddingUint8Test)
-ARMNN_AUTO_TEST_CASE(IgnorePaddingAveragePooling2dSize3, IgnorePaddingAveragePooling2dSize3Test)
-ARMNN_AUTO_TEST_CASE(IgnorePaddingAveragePooling2dSize3Uint8, IgnorePaddingAveragePooling2dSize3Uint8Test)
-
-ARMNN_AUTO_TEST_CASE(IgnorePaddingSimpleL2Pooling2d, IgnorePaddingSimpleL2Pooling2dTest)
-ARMNN_AUTO_TEST_CASE(UNSUPPORTED_IgnorePaddingSimpleL2Pooling2dUint8, IgnorePaddingSimpleL2Pooling2dUint8Test)
-ARMNN_AUTO_TEST_CASE(IgnorePaddingL2Pooling2dSize3, IgnorePaddingL2Pooling2dSize3Test)
-ARMNN_AUTO_TEST_CASE(UNSUPPORTED_IgnorePaddingL2Pooling2dSize3Uint8, IgnorePaddingL2Pooling2dSize3Uint8Test)
-
-ARMNN_AUTO_TEST_CASE(SimpleAveragePooling2d, SimpleAveragePooling2dTest)
-ARMNN_AUTO_TEST_CASE(SimpleAveragePooling2dNhwc, SimpleAveragePooling2dNhwcTest)
-ARMNN_AUTO_TEST_CASE(SimpleAveragePooling2dUint8, SimpleAveragePooling2dUint8Test)
-ARMNN_AUTO_TEST_CASE(IgnorePaddingAveragePooling2dSize3x2Stride2x2,
-                     IgnorePaddingAveragePooling2dSize3x2Stride2x2Test,
-                     false)
-ARMNN_AUTO_TEST_CASE(IgnorePaddingAveragePooling2dSize3x2Stride2x2NoPadding,
-                     IgnorePaddingAveragePooling2dSize3x2Stride2x2Test,
-                     true)
-ARMNN_AUTO_TEST_CASE(LargeTensorsAveragePooling2d, LargeTensorsAveragePooling2dTest)
-ARMNN_AUTO_TEST_CASE(LargeTensorsAveragePooling2dUint8, LargeTensorsAveragePooling2dUint8Test)
-
-ARMNN_AUTO_TEST_CASE(SimpleL2Pooling2d, SimpleL2Pooling2dTest)
-ARMNN_AUTO_TEST_CASE(UNSUPPORTED_SimpleL2Pooling2dUint8, SimpleL2Pooling2dUint8Test)
-ARMNN_AUTO_TEST_CASE(L2Pooling2dSize3Stride1, L2Pooling2dSize3Stride1Test)
-ARMNN_AUTO_TEST_CASE(UNSUPPORTED_L2Pooling2dSize3Stride1Uint8, L2Pooling2dSize3Stride1Uint8Test)
-ARMNN_AUTO_TEST_CASE(L2Pooling2dSize3Stride3, L2Pooling2dSize3Stride3Test)
-ARMNN_AUTO_TEST_CASE(UNSUPPORTED_L2Pooling2dSize3Stride3Uint8, L2Pooling2dSize3Stride3Uint8Test)
-ARMNN_AUTO_TEST_CASE(L2Pooling2dSize3Stride4, L2Pooling2dSize3Stride4Test)
-ARMNN_AUTO_TEST_CASE(UNSUPPORTED_L2Pooling2dSize3Stride4Uint8, L2Pooling2dSize3Stride4Uint8Test)
-ARMNN_AUTO_TEST_CASE(L2Pooling2dSize7, L2Pooling2dSize7Test)
-ARMNN_AUTO_TEST_CASE(UNSUPPORTED_L2Pooling2dSize7Uint8, L2Pooling2dSize7Uint8Test)
-ARMNN_AUTO_TEST_CASE(L2Pooling2dSize9, L2Pooling2dSize9Test)
-ARMNN_AUTO_TEST_CASE(UNSUPPORTED_L2Pooling2dSize9Uint8, L2Pooling2dSize9Uint8Test)
-
-// Add
-ARMNN_AUTO_TEST_CASE(SimpleAdd, AdditionTest)
-ARMNN_AUTO_TEST_CASE(AddBroadcast1Element, AdditionBroadcast1ElementTest)
-ARMNN_AUTO_TEST_CASE(AddBroadcast, AdditionBroadcastTest)
-
-ARMNN_AUTO_TEST_CASE(AdditionUint8, AdditionUint8Test)
-ARMNN_AUTO_TEST_CASE(AddBroadcastUint8, AdditionBroadcastUint8Test)
-ARMNN_AUTO_TEST_CASE(AddBroadcast1ElementUint8, AdditionBroadcast1ElementUint8Test)
-
-// Sub
-ARMNN_AUTO_TEST_CASE(SimpleSub, SubtractionTest)
-
-// Div
-ARMNN_AUTO_TEST_CASE(SimpleDivision, DivisionTest)
-ARMNN_AUTO_TEST_CASE(DivisionByZero, DivisionByZeroTest)
-ARMNN_AUTO_TEST_CASE(DivisionBroadcast1Element, DivisionBroadcast1ElementTest)
-ARMNN_AUTO_TEST_CASE(DivisionBroadcast1DVector, DivisionBroadcast1DVectorTest)
-// NOTE: quantized division is not supported by CL and not required by the
-//       android NN api
-
-// Mul
-ARMNN_AUTO_TEST_CASE(SimpleMultiplication, MultiplicationTest)
-ARMNN_AUTO_TEST_CASE(MultiplicationBroadcast1Element, MultiplicationBroadcast1ElementTest)
-ARMNN_AUTO_TEST_CASE(MultiplicationBroadcast1DVector, MultiplicationBroadcast1DVectorTest)
-ARMNN_AUTO_TEST_CASE(MultiplicationUint8, MultiplicationUint8Test)
-ARMNN_AUTO_TEST_CASE(MultiplicationBroadcast1ElementUint8, MultiplicationBroadcast1ElementUint8Test)
-ARMNN_AUTO_TEST_CASE(MultiplicationBroadcast1DVectorUint8, MultiplicationBroadcast1DVectorUint8Test)
-
-// Batch Norm
-ARMNN_AUTO_TEST_CASE(BatchNorm, BatchNormTest)
-
-// L2 Normalization
-ARMNN_AUTO_TEST_CASE(L2Normalization1d, L2Normalization1dTest)
-ARMNN_AUTO_TEST_CASE(L2Normalization2d, L2Normalization2dTest)
-ARMNN_AUTO_TEST_CASE(L2Normalization3d, L2Normalization3dTest)
-ARMNN_AUTO_TEST_CASE(L2Normalization4d, L2Normalization4dTest)
-
-ARMNN_AUTO_TEST_CASE(L2Normalization1dNhwc, L2Normalization1dNhwcTest)
-ARMNN_AUTO_TEST_CASE(L2Normalization2dNhwc, L2Normalization2dNhwcTest)
-ARMNN_AUTO_TEST_CASE(L2Normalization3dNhwc, L2Normalization3dNhwcTest)
-ARMNN_AUTO_TEST_CASE(L2Normalization4dNhwc, L2Normalization4dNhwcTest)
-
-// Resize Bilinear - NCHW data layout
-ARMNN_AUTO_TEST_CASE(ResizeBilinearNop, ResizeBilinearNopTest)
-ARMNN_AUTO_TEST_CASE(SimpleResizeBilinear, SimpleResizeBilinearTest)
-ARMNN_AUTO_TEST_CASE(ResizeBilinearSqMin, ResizeBilinearSqMinTest)
-ARMNN_AUTO_TEST_CASE(ResizeBilinearMin, ResizeBilinearMinTest)
-ARMNN_AUTO_TEST_CASE(ResizeBilinearMag, ResizeBilinearMagTest)
-
-// Resize Bilinear - NHWC data layout
-ARMNN_AUTO_TEST_CASE(ResizeBilinearNopNhwc, ResizeBilinearNopNhwcTest)
-ARMNN_AUTO_TEST_CASE(SimpleResizeBilinearNhwc, SimpleResizeBilinearNhwcTest)
-ARMNN_AUTO_TEST_CASE(ResizeBilinearSqMinNhwc, ResizeBilinearSqMinNhwcTest)
-ARMNN_AUTO_TEST_CASE(ResizeBilinearMinNhwc, ResizeBilinearMinNhwcTest)
-ARMNN_AUTO_TEST_CASE(ResizeBilinearMagNhwc, ResizeBilinearMagNhwcTest)
-
-// Constant
-ARMNN_AUTO_TEST_CASE(Constant, ConstantTest)
-ARMNN_AUTO_TEST_CASE(ConstantUint8, ConstantTestUint8)
-
-// Concat
-ARMNN_AUTO_TEST_CASE(Concatenation1d, Concatenation1dTest)
-ARMNN_AUTO_TEST_CASE(Concatenation1dUint8, Concatenation1dUint8Test)
-
-ARMNN_AUTO_TEST_CASE(Concatenation2dDim0, Concatenation2dDim0Test)
-ARMNN_AUTO_TEST_CASE(Concatenation2dDim0Uint8, Concatenation2dDim0Uint8Test)
-ARMNN_AUTO_TEST_CASE(Concatenation2dDim1, Concatenation2dDim1Test)
-ARMNN_AUTO_TEST_CASE(Concatenation2dDim1Uint8, Concatenation2dDim1Uint8Test)
-
-ARMNN_AUTO_TEST_CASE(Concatenation2dDim0DiffInputDims, Concatenation2dDim0DiffInputDimsTest)
-ARMNN_AUTO_TEST_CASE(Concatenation2dDim0DiffInputDimsUint8, Concatenation2dDim0DiffInputDimsUint8Test)
-ARMNN_AUTO_TEST_CASE(Concatenation2dDim1DiffInputDims, Concatenation2dDim1DiffInputDimsTest)
-ARMNN_AUTO_TEST_CASE(Concatenation2dDim1DiffInputDimsUint8, Concatenation2dDim1DiffInputDimsUint8Test)
-
-ARMNN_AUTO_TEST_CASE(Concatenation3dDim0, Concatenation3dDim0Test)
-ARMNN_AUTO_TEST_CASE(Concatenation3dDim0Uint8, Concatenation3dDim0Uint8Test)
-ARMNN_AUTO_TEST_CASE(Concatenation3dDim1, Concatenation3dDim1Test)
-ARMNN_AUTO_TEST_CASE(Concatenation3dDim1Uint8, Concatenation3dDim1Uint8Test)
-ARMNN_AUTO_TEST_CASE(Concatenation3dDim2, Concatenation3dDim2Test)
-ARMNN_AUTO_TEST_CASE(Concatenation3dDim2Uint8, Concatenation3dDim2Uint8Test)
-
-ARMNN_AUTO_TEST_CASE(Concatenation3dDim0DiffInputDims, Concatenation3dDim0DiffInputDimsTest)
-ARMNN_AUTO_TEST_CASE(Concatenation3dDim0DiffInputDimsUint8, Concatenation3dDim0DiffInputDimsUint8Test)
-ARMNN_AUTO_TEST_CASE(Concatenation3dDim1DiffInputDims, Concatenation3dDim1DiffInputDimsTest)
-ARMNN_AUTO_TEST_CASE(Concatenation3dDim1DiffInputDimsUint8, Concatenation3dDim1DiffInputDimsUint8Test)
-ARMNN_AUTO_TEST_CASE(Concatenation3dDim2DiffInputDims, Concatenation3dDim2DiffInputDimsTest)
-ARMNN_AUTO_TEST_CASE(Concatenation3dDim2DiffInputDimsUint8, Concatenation3dDim2DiffInputDimsUint8Test)
-
-// Floor
-ARMNN_AUTO_TEST_CASE(SimpleFloor, SimpleFloorTest)
-
-// Reshape
-ARMNN_AUTO_TEST_CASE(SimpleReshapeFloat32, SimpleReshapeFloat32Test)
-ARMNN_AUTO_TEST_CASE(SimpleReshapeUint8, SimpleReshapeUint8Test)
-
-// Permute
-ARMNN_AUTO_TEST_CASE(SimplePermuteFloat32, SimplePermuteFloat32Test)
-ARMNN_AUTO_TEST_CASE(SimplePermuteUint8, SimplePermuteUint8Test)
-ARMNN_AUTO_TEST_CASE(PermuteFloat32ValueSet1, PermuteFloat32ValueSet1Test)
-ARMNN_AUTO_TEST_CASE(PermuteFloat32ValueSet2, PermuteFloat32ValueSet2Test)
-ARMNN_AUTO_TEST_CASE(PermuteFloat32ValueSet3, PermuteFloat32ValueSet3Test)
-
-// Lstm
-ARMNN_AUTO_TEST_CASE(LstmLayerFloat32WithCifgWithPeepholeNoProjection,
-                     LstmLayerFloat32WithCifgWithPeepholeNoProjectionTest)
-ARMNN_AUTO_TEST_CASE(LstmLayerFloat32NoCifgNoPeepholeNoProjection,
-                     LstmLayerFloat32NoCifgNoPeepholeNoProjectionTest)
-ARMNN_AUTO_TEST_CASE(LstmLayerFloat32NoCifgWithPeepholeWithProjection,
-                     LstmLayerFloat32NoCifgWithPeepholeWithProjectionTest)
-
-// Convert from Float16 to Float32
-ARMNN_AUTO_TEST_CASE(SimpleConvertFp16ToFp32, SimpleConvertFp16ToFp32Test)
-// Convert from Float32 to Float16
-ARMNN_AUTO_TEST_CASE(SimpleConvertFp32ToFp16, SimpleConvertFp32ToFp16Test)
-
-// ============================================================================
-// COMPARE tests
-
-ARMNN_COMPARE_REF_AUTO_TEST_CASE(CompareConv2dWithReference, CompareConvolution2dTest)
-
-ARMNN_COMPARE_REF_AUTO_TEST_CASE(CompareDepthwiseConv2dWithReferenceFloat32, CompareDepthwiseConvolution2dTest<float>)
-ARMNN_COMPARE_REF_AUTO_TEST_CASE(CompareDepthwiseConv2dWithReferenceUint8, CompareDepthwiseConvolution2dTest<uint8_t>)
-
-ARMNN_COMPARE_REF_AUTO_TEST_CASE(CompareNormalizationWithinWithReference, CompareNormalizationTest,
-                                 armnn::NormalizationAlgorithmChannel::Within,
-                                 armnn::NormalizationAlgorithmMethod::LocalBrightness)
-ARMNN_COMPARE_REF_AUTO_TEST_CASE(CompareNormalizationAcrossWithReference, CompareNormalizationTest,
-                                 armnn::NormalizationAlgorithmChannel::Across,
-                                 armnn::NormalizationAlgorithmMethod::LocalBrightness)
-
-ARMNN_COMPARE_REF_AUTO_TEST_CASE(CompareSoftmaxBeta1WithReference, CompareSoftmaxTest, 1.0f)
-ARMNN_COMPARE_REF_AUTO_TEST_CASE(CompareSoftmaxBeta2WithReference, CompareSoftmaxTest, 2.0f)
-ARMNN_COMPARE_REF_AUTO_TEST_CASE(CompareSoftmaxUint8, CompareSoftmaxUint8Test, 1.0f)
-
-ARMNN_COMPARE_REF_AUTO_TEST_CASE(CompareMaxPooling2dWithRef, ComparePooling2dTest, armnn::PoolingAlgorithm::Max)
-
-ARMNN_COMPARE_REF_AUTO_TEST_CASE(CompareAveragePooling2dWithRef, ComparePooling2dTest, armnn::PoolingAlgorithm::Average)
-ARMNN_COMPARE_REF_AUTO_TEST_CASE(CompareAveragePooling2dWithRefUint8, ComparePooling2dUint8Test,
-                                 armnn::PoolingAlgorithm::Average)
-
-ARMNN_COMPARE_REF_AUTO_TEST_CASE(CompareL2Pooling2dWithRef, ComparePooling2dTest, armnn::PoolingAlgorithm::L2)
-
-ARMNN_COMPARE_REF_AUTO_TEST_CASE(CompareAddition, CompareAdditionTest)
-
-ARMNN_COMPARE_REF_AUTO_TEST_CASE(CompareMultiplicationWithRef, CompareMultiplicationTest)
-
-ARMNN_COMPARE_REF_AUTO_TEST_CASE(CompareBatchNorm, CompareBatchNormTest)
-
-ARMNN_COMPARE_REF_AUTO_TEST_CASE(CompareReLu1, CompareBoundedReLuTest, 1.0f, -1.0f)
-ARMNN_COMPARE_REF_AUTO_TEST_CASE(CompareReLu6, CompareBoundedReLuTest, 6.0f, 0.0f)
-
-// ============================================================================
-// FIXTURE tests
-
-ARMNN_COMPARE_REF_FIXTURE_TEST_CASE(CompareSigmoidActivationWithReference, ActivationFixture,
-                                    CompareActivationTest, armnn::ActivationFunction::Sigmoid, 5u)
-
-ARMNN_COMPARE_REF_FIXTURE_TEST_CASE(CompareTanhActivationWithReference, ActivationFixture,
-                                    CompareActivationTest, armnn::ActivationFunction::TanH, 5u)
-
-ARMNN_COMPARE_REF_FIXTURE_TEST_CASE(CompareLinearActivationWithReference, ActivationFixture,
-                                    CompareActivationTest, armnn::ActivationFunction::Linear, 5u)
-
-ARMNN_COMPARE_REF_FIXTURE_TEST_CASE(CompareReLuActivationWithReference, ActivationFixture,
-                                    CompareActivationTest, armnn::ActivationFunction::ReLu, 5u)
-
-ARMNN_COMPARE_REF_FIXTURE_TEST_CASE(CompareBoundedReLuActivationWithReference, ActivationFixture,
-                                    CompareActivationTest, armnn::ActivationFunction::BoundedReLu, 5u)
-ARMNN_COMPARE_REF_FIXTURE_TEST_CASE(CompareBoundedReLuActivationWithReferenceUint8, ActivationFixture,
-                                    CompareActivationUint8Test, armnn::ActivationFunction::BoundedReLu)
-
-ARMNN_COMPARE_REF_FIXTURE_TEST_CASE(CompareSoftReLuActivationWithReference, ActivationFixture,
-                                    CompareActivationTest, armnn::ActivationFunction::SoftReLu, 5u)
-
-ARMNN_COMPARE_REF_FIXTURE_TEST_CASE(CompareLeakyReLuActivationWithReference, ActivationFixture,
-                                    CompareActivationTest, armnn::ActivationFunction::LeakyReLu, 5u)
-
-ARMNN_COMPARE_REF_FIXTURE_TEST_CASE(CompareAbsActivationWithReference, ActivationFixture,
-                                    CompareActivationTest, armnn::ActivationFunction::Abs, 5u)
-
-ARMNN_COMPARE_REF_FIXTURE_TEST_CASE(CompareSqrtActivationWithReference, PositiveActivationFixture,
-                                    CompareActivationTest, armnn::ActivationFunction::Sqrt, 5u)
-
-ARMNN_COMPARE_REF_FIXTURE_TEST_CASE(CompareSquareActivationWithReference, ActivationFixture,
-                                    CompareActivationTest, armnn::ActivationFunction::Square, 5u)
-
-BOOST_AUTO_TEST_SUITE_END()
diff --git a/src/backends/test/ArmComputeNeon.cpp b/src/backends/test/ArmComputeNeon.cpp
deleted file mode 100644
index 045aa30889..0000000000
--- a/src/backends/test/ArmComputeNeon.cpp
+++ /dev/null
@@ -1,485 +0,0 @@
-//
-// Copyright © 2017 Arm Ltd. All rights reserved.
-// SPDX-License-Identifier: MIT
-//
-#include <boost/test/unit_test.hpp>
-
-#include "test/TensorHelpers.hpp"
-#include "LayerTests.hpp"
-
-#include <backends/CpuTensorHandle.hpp>
-#include <backends/neon/NeonLayerSupport.hpp>
-#include <backends/neon/NeonWorkloadFactory.hpp>
-#include <backends/reference/RefWorkloadFactory.hpp>
-#include <backends/test/TensorCopyUtils.hpp>
-#include "ActivationFixture.hpp"
-
-#include "WorkloadTestUtils.hpp"
-
-#include "test/UnitTests.hpp"
-
-BOOST_AUTO_TEST_SUITE(Compute_ArmComputeNeon)
-using FactoryType = armnn::NeonWorkloadFactory;
-
-// ============================================================================
-// UNIT tests
-
-// Convolution
-ARMNN_AUTO_TEST_CASE(SimpleConvolution1d, Convolution1dTest, true)
-
-ARMNN_AUTO_TEST_CASE(SimpleConvolution2d, SimpleConvolution2d3x5Test, true)
-ARMNN_AUTO_TEST_CASE(SimpleConvolution2dSquare, SimpleConvolution2d3x3Test, true)
-ARMNN_AUTO_TEST_CASE(UnbiasedConvolution2d, SimpleConvolution2d3x5Test, false)
-ARMNN_AUTO_TEST_CASE(UnbiasedConvolution2dSquare, SimpleConvolution2d3x3Test, false)
-ARMNN_AUTO_TEST_CASE(SimpleConvolution2dAsymmetricPadding, Convolution2dAsymmetricPaddingTest)
-
-ARMNN_AUTO_TEST_CASE(SimpleConvolution2dSquareNhwc, SimpleConvolution2d3x3NhwcTest, false)
-namespace
-{
-
-armnn::Convolution2dDescriptor MakeConv2dDesc(uint32_t strideX, uint32_t strideY,
-    uint32_t padLeft = 0, uint32_t padRight = 0, uint32_t padTop = 0, uint32_t padBottom = 0)
-{
-    armnn::Convolution2dDescriptor result;
-    result.m_StrideX = strideX;
-    result.m_StrideY = strideY;
-    result.m_PadLeft = padLeft;
-    result.m_PadRight = padRight;
-    result.m_PadTop = padTop;
-    result.m_PadBottom = padBottom;
-    result.m_BiasEnabled = true;
-    return result;
-}
-
-}
-
-BOOST_AUTO_TEST_CASE(Conv2dUtils)
-{
-    // The only preferred Neon convolution is 1x1 with padding=0 and stride size {1,2,3}.
-    armnn::TensorShape shape1x1({ 1,1,1,1 });
-    armnn::TensorInfo info1x1(shape1x1, armnn::DataType::Float32);
-    BOOST_TEST(armnn::IsNeonDirectConvolutionPreferred(info1x1, MakeConv2dDesc(1, 1)));
-    BOOST_TEST(armnn::IsNeonDirectConvolutionPreferred(info1x1, MakeConv2dDesc(1, 2)));
-    BOOST_TEST(armnn::IsNeonDirectConvolutionPreferred(info1x1, MakeConv2dDesc(1, 3)));
-    BOOST_TEST(armnn::IsNeonDirectConvolutionPreferred(info1x1, MakeConv2dDesc(2, 1)));
-    BOOST_TEST(armnn::IsNeonDirectConvolutionPreferred(info1x1, MakeConv2dDesc(2, 2)));
-    BOOST_TEST(armnn::IsNeonDirectConvolutionPreferred(info1x1, MakeConv2dDesc(2, 3)));
-    BOOST_TEST(armnn::IsNeonDirectConvolutionPreferred(info1x1, MakeConv2dDesc(3, 1)));
-    BOOST_TEST(armnn::IsNeonDirectConvolutionPreferred(info1x1, MakeConv2dDesc(3, 2)));
-    BOOST_TEST(armnn::IsNeonDirectConvolutionPreferred(info1x1, MakeConv2dDesc(3, 3)));
-
-    BOOST_TEST(!armnn::IsNeonDirectConvolutionPreferred(info1x1, MakeConv2dDesc(4, 1)));
-    BOOST_TEST(!armnn::IsNeonDirectConvolutionPreferred(info1x1, MakeConv2dDesc(4, 5)));
-    BOOST_TEST(!armnn::IsNeonDirectConvolutionPreferred(info1x1, MakeConv2dDesc(3, 6)));
-
-    // non zero padding is not preferred for direct convolution
-    BOOST_TEST(!armnn::IsNeonDirectConvolutionPreferred(info1x1, MakeConv2dDesc(1, 1, 1, 0)));
-    BOOST_TEST(!armnn::IsNeonDirectConvolutionPreferred(info1x1, MakeConv2dDesc(1, 1, 0, 1)));
-    BOOST_TEST(!armnn::IsNeonDirectConvolutionPreferred(info1x1, MakeConv2dDesc(1, 1, 1, 1)));
-
-    // 2x2 filter not preferred for direct convolution
-    armnn::TensorShape shape2x2({ 1,1,2,2 });
-    armnn::TensorInfo info2x2(shape2x2, armnn::DataType::Float32);
-    BOOST_TEST(!armnn::IsNeonDirectConvolutionPreferred(info2x2, MakeConv2dDesc(1, 1)));
-}
-
-// Depthwise Convolution
-ARMNN_AUTO_TEST_CASE(DepthwiseConvolution2dDepthMul1, DepthwiseConvolution2dDepthMul1Test, true)
-ARMNN_AUTO_TEST_CASE(UnbiasedDepthwiseConvolution2dDepthMul1, DepthwiseConvolution2dDepthMul1Test, false)
-ARMNN_AUTO_TEST_CASE(DepthwiseConvolution2dDepthMul1Uint8, DepthwiseConvolution2dDepthMul1Uint8Test, true)
-ARMNN_AUTO_TEST_CASE(UnbiasedDepthwiseConvolution2dDepthMul1Uint8, DepthwiseConvolution2dDepthMul1Uint8Test, false)
-
-ARMNN_AUTO_TEST_CASE(DepthwiseConvolution2dAsymmetric, DepthwiseConvolution2dAsymmetricTest, true)
-ARMNN_AUTO_TEST_CASE(UnbiasedDepthwiseConvolution2dAsymmetric, DepthwiseConvolution2dAsymmetricTest, false)
-
-namespace
-{
-
-armnn::DepthwiseConvolution2dDescriptor MakeDepthwiseConv2dDesc(uint32_t strideX, uint32_t strideY,
-    uint32_t depthMultiplier = 1, uint32_t padLeft = 0, uint32_t padRight = 0,
-    uint32_t padTop = 0, uint32_t padBottom = 0)
-{
-    boost::ignore_unused(depthMultiplier);
-
-    armnn::DepthwiseConvolution2dDescriptor desc;
-
-    desc.m_PadLeft = padLeft;
-    desc.m_PadRight = padRight;
-
-    desc.m_PadTop = padTop;
-    desc.m_PadBottom = padBottom;
-    desc.m_StrideX = strideX;
-    desc.m_StrideY = strideY;
-    desc.m_BiasEnabled = false;
-
-    return desc;
-}
-
-armnn::TensorInfo CreateOutputTensorInfo(const armnn::TensorInfo& inputInfo,
-                                         const armnn::TensorInfo& weightsInfo,
-                                         const armnn::DepthwiseConvolution2dDescriptor& descriptor,
-                                         armnn::DataType dataType)
-{
-    const armnn::TensorShape& inputShape  = inputInfo.GetShape();
-    const armnn::TensorShape& filterShape = weightsInfo.GetShape();
-
-    unsigned int inWidth = inputShape[3];
-    unsigned int inHeight = inputShape[2];
-    unsigned int inBatchSize = inputShape[0];
-
-    unsigned int filterWidth = filterShape[3];
-    unsigned int readWidth = (inWidth + descriptor.m_PadLeft + descriptor.m_PadRight) - (filterWidth);
-    unsigned int outWidth =  1u + (readWidth / descriptor.m_StrideX);
-
-    unsigned int filterHeight = filterShape[2];
-    unsigned int readHeight = (inHeight + descriptor.m_PadTop + descriptor.m_PadBottom) - (filterHeight);
-    unsigned int outHeight = 1u + (readHeight / descriptor.m_StrideY);
-    unsigned int depthMultiplier = filterShape[0];
-
-    unsigned int outChannels = filterShape[1] * depthMultiplier;
-    unsigned int outBatchSize = inBatchSize;
-
-    armnn::TensorShape outputShape({outBatchSize, outChannels, outHeight, outWidth});
-    return armnn::TensorInfo(outputShape, dataType);
-}
-}
-
-BOOST_AUTO_TEST_CASE(DepthwiseConv2dUtils)
-{
-    const armnn::DataType dataType = armnn::DataType::Float32;
-
-    armnn::TensorInfo inputInfo({1, 1, 10, 10 }, dataType);
-    armnn::TensorInfo outputInfo;
-    armnn::TensorInfo weightsInfo3x3({ 1, 1, 3, 3 }, dataType);
-    armnn::TensorInfo biasesInfo;
-
-    armnn::DepthwiseConvolution2dDescriptor descriptor;
-
-    // Strides supported: 1,2,3
-    descriptor = MakeDepthwiseConv2dDesc(1, 1);
-    outputInfo = CreateOutputTensorInfo(inputInfo, weightsInfo3x3, descriptor, dataType);
-    BOOST_TEST(armnn::IsDepthwiseConvolutionSupportedNeon(inputInfo, outputInfo, descriptor,
-                                                          weightsInfo3x3, biasesInfo));
-
-    descriptor = MakeDepthwiseConv2dDesc(1, 2);
-    outputInfo = CreateOutputTensorInfo(inputInfo, weightsInfo3x3, descriptor, dataType);
-    BOOST_TEST(armnn::IsDepthwiseConvolutionSupportedNeon(inputInfo, outputInfo, descriptor,
-                                                          weightsInfo3x3, biasesInfo));
-
-    descriptor = MakeDepthwiseConv2dDesc(1, 3);
-    outputInfo = CreateOutputTensorInfo(inputInfo, weightsInfo3x3, descriptor, dataType);
-    BOOST_TEST(armnn::IsDepthwiseConvolutionSupportedNeon(inputInfo, outputInfo, descriptor,
-                                                          weightsInfo3x3, biasesInfo));
-
-    descriptor = MakeDepthwiseConv2dDesc(2, 1);
-    outputInfo = CreateOutputTensorInfo(inputInfo, weightsInfo3x3, descriptor, dataType);
-    BOOST_TEST(armnn::IsDepthwiseConvolutionSupportedNeon(inputInfo, outputInfo, descriptor,
-                                                          weightsInfo3x3, biasesInfo));
-
-    descriptor = MakeDepthwiseConv2dDesc(2, 2);
-    outputInfo = CreateOutputTensorInfo(inputInfo, weightsInfo3x3, descriptor, dataType);
-    BOOST_TEST(armnn::IsDepthwiseConvolutionSupportedNeon(inputInfo, outputInfo, descriptor,
-                                                          weightsInfo3x3, biasesInfo));
-
-    descriptor = MakeDepthwiseConv2dDesc(2, 3);
-    outputInfo = CreateOutputTensorInfo(inputInfo, weightsInfo3x3, descriptor, dataType);
-    BOOST_TEST(armnn::IsDepthwiseConvolutionSupportedNeon(inputInfo, outputInfo, descriptor,
-                                                          weightsInfo3x3, biasesInfo));
-
-    descriptor = MakeDepthwiseConv2dDesc(3, 1);
-    outputInfo = CreateOutputTensorInfo(inputInfo, weightsInfo3x3, descriptor, dataType);
-    BOOST_TEST(armnn::IsDepthwiseConvolutionSupportedNeon(inputInfo, outputInfo, descriptor,
-                                                          weightsInfo3x3, biasesInfo));
-
-    descriptor = MakeDepthwiseConv2dDesc(3, 2);
-    outputInfo = CreateOutputTensorInfo(inputInfo, weightsInfo3x3, descriptor, dataType);
-    BOOST_TEST(armnn::IsDepthwiseConvolutionSupportedNeon(inputInfo, outputInfo, descriptor,
-                                                          weightsInfo3x3, biasesInfo));
-
-    descriptor = MakeDepthwiseConv2dDesc(3, 3);
-    outputInfo = CreateOutputTensorInfo(inputInfo, weightsInfo3x3, descriptor, dataType);
-    BOOST_TEST(armnn::IsDepthwiseConvolutionSupportedNeon(inputInfo, outputInfo, descriptor,
-                                                          weightsInfo3x3, biasesInfo));
-
-    // Supported stride 4
-    descriptor = MakeDepthwiseConv2dDesc(4, 1);
-    outputInfo = CreateOutputTensorInfo(inputInfo, weightsInfo3x3, descriptor, dataType);
-    BOOST_TEST(armnn::IsDepthwiseConvolutionSupportedNeon(inputInfo, outputInfo, descriptor,
-                                                          weightsInfo3x3, biasesInfo));
-
-    // Supported weights shape 1x1
-    armnn::TensorInfo weightsInfo1x1({ 1, 1, 1, 1 }, armnn::DataType::Float32);
-    descriptor = MakeDepthwiseConv2dDesc(1, 1);
-    outputInfo = CreateOutputTensorInfo(inputInfo, weightsInfo1x1, descriptor, dataType);
-    BOOST_TEST(armnn::IsDepthwiseConvolutionSupportedNeon(inputInfo, outputInfo, descriptor,
-                                                          weightsInfo1x1, biasesInfo));
-
-    // Supported shape 2x2
-    armnn::TensorInfo weightsInfo2x2({ 1, 1, 2, 2 }, armnn::DataType::Float32);
-    descriptor = MakeDepthwiseConv2dDesc(1, 1);
-    outputInfo = CreateOutputTensorInfo(inputInfo, weightsInfo2x2, descriptor, dataType);
-    BOOST_TEST(armnn::IsDepthwiseConvolutionSupportedNeon(inputInfo, outputInfo, descriptor,
-                                                          weightsInfo2x2, biasesInfo));
-
-    // Asymmetric padding
-    descriptor = MakeDepthwiseConv2dDesc(1, 1, 1, 1, 2, 1, 2);
-    outputInfo = CreateOutputTensorInfo(inputInfo, weightsInfo3x3, descriptor, dataType);
-    BOOST_TEST(armnn::IsDepthwiseConvolutionSupportedNeon(inputInfo, outputInfo, descriptor,
-                                                          weightsInfo3x3, biasesInfo));
-}
-
-// Pooling
-ARMNN_AUTO_TEST_CASE(SimpleMaxPooling2dSize3x3Stride2x4, SimpleMaxPooling2dSize3x3Stride2x4Test, true)
-ARMNN_AUTO_TEST_CASE(SimpleMaxPooling2dSize3x3Stride2x4Uint8, SimpleMaxPooling2dSize3x3Stride2x4Uint8Test, true)
-ARMNN_AUTO_TEST_CASE(SimpleAveragePooling2d, SimpleAveragePooling2dTest)
-ARMNN_AUTO_TEST_CASE(SimpleAveragePooling2dNhwc, SimpleAveragePooling2dNhwcTest)
-ARMNN_AUTO_TEST_CASE(SimpleAveragePooling2dUint8, SimpleAveragePooling2dUint8Test)
-
-ARMNN_AUTO_TEST_CASE(LargeTensorsAveragePooling2d, LargeTensorsAveragePooling2dTest)
-ARMNN_AUTO_TEST_CASE(LargeTensorsAveragePooling2dUint8, LargeTensorsAveragePooling2dUint8Test)
-
-ARMNN_AUTO_TEST_CASE(SimpleL2Pooling2d, SimpleL2Pooling2dTest)
-ARMNN_AUTO_TEST_CASE(UNSUPPORTED_SimpleL2Pooling2dUint8, SimpleL2Pooling2dUint8Test)
-ARMNN_AUTO_TEST_CASE(L2Pooling2dSize3Stride1, L2Pooling2dSize3Stride1Test)
-ARMNN_AUTO_TEST_CASE(UNSUPPORTED_L2Pooling2dSize3Stride1Uint8, L2Pooling2dSize3Stride1Uint8Test)
-ARMNN_AUTO_TEST_CASE(L2Pooling2dSize3Stride3, L2Pooling2dSize3Stride3Test)
-ARMNN_AUTO_TEST_CASE(UNSUPPORTED_L2Pooling2dSize3Stride3Uint8, L2Pooling2dSize3Stride3Uint8Test)
-ARMNN_AUTO_TEST_CASE(L2Pooling2dSize3Stride4, L2Pooling2dSize3Stride4Test)
-ARMNN_AUTO_TEST_CASE(UNSUPPORTED_L2Pooling2dSize3Stride4Uint8, L2Pooling2dSize3Stride4Uint8Test)
-ARMNN_AUTO_TEST_CASE(L2Pooling2dSize7, L2Pooling2dSize7Test)
-ARMNN_AUTO_TEST_CASE(UNSUPPORTED_L2Pooling2dSize7Uint8, L2Pooling2dSize7Uint8Test)
-ARMNN_AUTO_TEST_CASE(L2Pooling2dSize9, L2Pooling2dSize9Test)
-ARMNN_AUTO_TEST_CASE(UNSUPPORTED_L2Pooling2dSize9Uint8, L2Pooling2dSize9Uint8Test)
-
-// Ignore padding values for pooling but count padding fields into the divisor
-ARMNN_AUTO_TEST_CASE(IgnorePaddingSimpleMaxPooling2d, IgnorePaddingSimpleMaxPooling2dTest)
-ARMNN_AUTO_TEST_CASE(IgnorePaddingSimpleMaxPooling2dUint8, IgnorePaddingSimpleMaxPooling2dUint8Test)
-ARMNN_AUTO_TEST_CASE(IgnorePaddingMaxPooling2dSize3, IgnorePaddingMaxPooling2dSize3Test)
-ARMNN_AUTO_TEST_CASE(IgnorePaddingMaxPooling2dSize3Uint8, IgnorePaddingMaxPooling2dSize3Uint8Test)
-
-ARMNN_AUTO_TEST_CASE(IgnorePaddingSimpleAveragePooling2d, IgnorePaddingSimpleAveragePooling2dTest)
-ARMNN_AUTO_TEST_CASE(IgnorePaddingSimpleAveragePooling2dUint8, IgnorePaddingSimpleAveragePooling2dUint8Test)
-ARMNN_AUTO_TEST_CASE(IgnorePaddingSimpleAveragePooling2dNoPadding, IgnorePaddingSimpleAveragePooling2dNoPaddingTest)
-ARMNN_AUTO_TEST_CASE(IgnorePaddingSimpleAveragePooling2dNoPaddingUint8,
-    IgnorePaddingSimpleAveragePooling2dNoPaddingUint8Test)
-ARMNN_AUTO_TEST_CASE(IgnorePaddingAveragePooling2dSize3, IgnorePaddingAveragePooling2dSize3Test)
-ARMNN_AUTO_TEST_CASE(IgnorePaddingAveragePooling2dSize3Uint8, IgnorePaddingAveragePooling2dSize3Uint8Test)
-ARMNN_AUTO_TEST_CASE(IgnorePaddingAveragePooling2dSize3x2Stride2x2,
-                             IgnorePaddingAveragePooling2dSize3x2Stride2x2Test, false)
-ARMNN_AUTO_TEST_CASE(IgnorePaddingAveragePooling2dSize3x2Stride2x2NoPadding,
-                             IgnorePaddingAveragePooling2dSize3x2Stride2x2Test,
-                                          true)
-
-ARMNN_AUTO_TEST_CASE(IgnorePaddingSimpleL2Pooling2d, IgnorePaddingSimpleL2Pooling2dTest)
-ARMNN_AUTO_TEST_CASE(UNSUPPORTED_IgnorePaddingSimpleL2Pooling2dUint8, IgnorePaddingSimpleL2Pooling2dUint8Test)
-ARMNN_AUTO_TEST_CASE(IgnorePaddingL2Pooling2dSize3, IgnorePaddingL2Pooling2dSize3Test)
-ARMNN_AUTO_TEST_CASE(UNSUPPORTED_IgnorePaddingL2Pooling2dSize3Uint8, IgnorePaddingL2Pooling2dSize3Uint8Test)
-
-// Activation
-ARMNN_AUTO_TEST_CASE(ConstantLinearActivation, ConstantLinearActivationTest)
-
-ARMNN_AUTO_TEST_CASE(SimpleSoftmaxBeta1, SimpleSoftmaxTest, 1.0f)
-ARMNN_AUTO_TEST_CASE(SimpleSoftmaxBeta2, SimpleSoftmaxTest, 2.0f)
-
-ARMNN_AUTO_TEST_CASE(SimpleSoftmaxBeta1Uint8, SimpleSoftmaxUint8Test, 1.0f)
-ARMNN_AUTO_TEST_CASE(SimpleSoftmaxBeta2Uint8, SimpleSoftmaxUint8Test, 2.0f)
-
-ARMNN_AUTO_TEST_CASE(ReLu1Uint8, BoundedReLuUint8UpperAndLowerBoundTest)
-ARMNN_AUTO_TEST_CASE(ReLu6Uint8, BoundedReLuUint8UpperBoundOnlyTest)
-
-// Softmax
-BOOST_AUTO_TEST_CASE(Softmax4dSupport)
-{
-    const unsigned int numDimensions = 4u;
-    std::array<unsigned int, numDimensions> dimensionSizes;
-    dimensionSizes.fill(1u);
-
-    const armnn::TensorInfo inputInfo(numDimensions, &dimensionSizes.front(), armnn::DataType::Float32);
-    const armnn::TensorInfo outputInfo(numDimensions, &dimensionSizes.front(), armnn::DataType::Float32);
-
-    // 4D Softmax should be reported as unsupported on the NEON backend
-    BOOST_TEST(!armnn::IsSoftmaxSupportedNeon(inputInfo, outputInfo, armnn::SoftmaxDescriptor()));
-}
-
-// Splitter
-ARMNN_AUTO_TEST_CASE(SimpleSplitter, SplitterTest)
-ARMNN_AUTO_TEST_CASE(SimpleSplitterUint8, SplitterUint8Test)
-
-ARMNN_AUTO_TEST_CASE(CopyViaSplitter, CopyViaSplitterTest)
-ARMNN_AUTO_TEST_CASE(CopyViaSplitterUint8, CopyViaSplitterUint8Test)
-
-// Merger
-ARMNN_AUTO_TEST_CASE(SimpleMerger, MergerTest)
-ARMNN_AUTO_TEST_CASE(MergerUint8, MergerUint8Test)
-
-// Fully Connected
-ARMNN_AUTO_TEST_CASE(SimpleFullyConnected, FullyConnectedFloat32Test, false, false)
-ARMNN_AUTO_TEST_CASE(SimpleFullyConnectedWithBias, FullyConnectedFloat32Test, true, false)
-ARMNN_AUTO_TEST_CASE(SimpleFullyConnectedWithTranspose, FullyConnectedFloat32Test, false, true)
-ARMNN_AUTO_TEST_CASE(FullyConnectedLarge, FullyConnectedLargeTest, false)
-ARMNN_AUTO_TEST_CASE(FullyConnectedLargeTransposed, FullyConnectedLargeTest, true)
-ARMNN_AUTO_TEST_CASE(FullyConnectedUint8, FullyConnectedUint8Test, false)
-ARMNN_AUTO_TEST_CASE(FullyConnectedBiasedUint8, FullyConnectedUint8Test, true)
-
-// Add
-ARMNN_AUTO_TEST_CASE(SimpleAdd, AdditionTest)
-ARMNN_AUTO_TEST_CASE(AddBroadcast, AdditionBroadcastTest)
-ARMNN_AUTO_TEST_CASE(AddBroadcast1Element, AdditionBroadcast1ElementTest)
-
-// Sub
-ARMNN_AUTO_TEST_CASE(SimpleSub, SubtractionTest)
-
-// Mul
-ARMNN_AUTO_TEST_CASE(SimpleMultiplication, MultiplicationTest)
-ARMNN_AUTO_TEST_CASE(MultiplicationBroadcast1Element, MultiplicationBroadcast1ElementTest)
-ARMNN_AUTO_TEST_CASE(MultiplicationBroadcast1DVector, MultiplicationBroadcast1DVectorTest)
-
-// Batch Norm
-ARMNN_AUTO_TEST_CASE(BatchNorm, BatchNormTest)
-
-// Constant
-ARMNN_AUTO_TEST_CASE(Constant, ConstantTest)
-ARMNN_AUTO_TEST_CASE(ConstantUint8, ConstantTestUint8)
-
-// Concatenation
-ARMNN_AUTO_TEST_CASE(Concatenation1d, Concatenation1dTest)
-ARMNN_AUTO_TEST_CASE(Concatenation1dUint8, Concatenation1dUint8Test)
-
-ARMNN_AUTO_TEST_CASE(Concatenation2dDim0, Concatenation2dDim0Test)
-ARMNN_AUTO_TEST_CASE(Concatenation2dDim0Uint8, Concatenation2dDim0Uint8Test)
-ARMNN_AUTO_TEST_CASE(Concatenation2dDim1, Concatenation2dDim1Test)
-ARMNN_AUTO_TEST_CASE(Concatenation2dDim1Uint8, Concatenation2dDim1Uint8Test)
-
-ARMNN_AUTO_TEST_CASE(Concatenation2dDim0DiffInputDims, Concatenation2dDim0DiffInputDimsTest)
-ARMNN_AUTO_TEST_CASE(Concatenation2dDim0DiffInputDimsUint8, Concatenation2dDim0DiffInputDimsUint8Test)
-ARMNN_AUTO_TEST_CASE(Concatenation2dDim1DiffInputDims, Concatenation2dDim1DiffInputDimsTest)
-ARMNN_AUTO_TEST_CASE(Concatenation2dDim1DiffInputDimsUint8, Concatenation2dDim1DiffInputDimsUint8Test)
-
-ARMNN_AUTO_TEST_CASE(Concatenation3dDim0, Concatenation3dDim0Test)
-ARMNN_AUTO_TEST_CASE(Concatenation3dDim0Uint8, Concatenation3dDim0Uint8Test)
-ARMNN_AUTO_TEST_CASE(Concatenation3dDim1, Concatenation3dDim1Test)
-ARMNN_AUTO_TEST_CASE(Concatenation3dDim1Uint8, Concatenation3dDim1Uint8Test)
-ARMNN_AUTO_TEST_CASE(Concatenation3dDim2, Concatenation3dDim2Test)
-ARMNN_AUTO_TEST_CASE(Concatenation3dDim2Uint8, Concatenation3dDim2Uint8Test)
-
-ARMNN_AUTO_TEST_CASE(Concatenation3dDim0DiffInputDims, Concatenation3dDim0DiffInputDimsTest)
-ARMNN_AUTO_TEST_CASE(Concatenation3dDim0DiffInputDimsUint8, Concatenation3dDim0DiffInputDimsUint8Test)
-ARMNN_AUTO_TEST_CASE(Concatenation3dDim1DiffInputDims, Concatenation3dDim1DiffInputDimsTest)
-ARMNN_AUTO_TEST_CASE(Concatenation3dDim1DiffInputDimsUint8, Concatenation3dDim1DiffInputDimsUint8Test)
-ARMNN_AUTO_TEST_CASE(Concatenation3dDim2DiffInputDims, Concatenation3dDim2DiffInputDimsTest)
-ARMNN_AUTO_TEST_CASE(Concatenation3dDim2DiffInputDimsUint8, Concatenation3dDim2DiffInputDimsUint8Test)
-
-// L2 Normalization
-ARMNN_AUTO_TEST_CASE(L2Normalization1d, L2Normalization1dTest)
-ARMNN_AUTO_TEST_CASE(L2Normalization2d, L2Normalization2dTest)
-ARMNN_AUTO_TEST_CASE(L2Normalization3d, L2Normalization3dTest)
-ARMNN_AUTO_TEST_CASE(L2Normalization4d, L2Normalization4dTest)
-
-ARMNN_AUTO_TEST_CASE(L2Normalization1dNhwc, L2Normalization1dNhwcTest)
-ARMNN_AUTO_TEST_CASE(L2Normalization2dNhwc, L2Normalization2dNhwcTest)
-ARMNN_AUTO_TEST_CASE(L2Normalization3dNhwc, L2Normalization3dNhwcTest)
-ARMNN_AUTO_TEST_CASE(L2Normalization4dNhwc, L2Normalization4dNhwcTest)
-
-// Floor
-ARMNN_AUTO_TEST_CASE(SimpleFloor, SimpleFloorTest)
-
-// Reshape
-ARMNN_AUTO_TEST_CASE(SimpleReshapeFloat32, SimpleReshapeFloat32Test)
-ARMNN_AUTO_TEST_CASE(SimpleReshapeUint8, SimpleReshapeUint8Test)
-
-// Permute
-ARMNN_AUTO_TEST_CASE(SimplePermuteFloat32, SimplePermuteFloat32Test)
-ARMNN_AUTO_TEST_CASE(SimplePermuteUint8, SimplePermuteUint8Test)
-ARMNN_AUTO_TEST_CASE(PermuteFloat32ValueSet1, PermuteFloat32ValueSet1Test)
-ARMNN_AUTO_TEST_CASE(PermuteFloat32ValueSet2, PermuteFloat32ValueSet2Test)
-ARMNN_AUTO_TEST_CASE(PermuteFloat32ValueSet3, PermuteFloat32ValueSet3Test)
-
-// Lstm
-ARMNN_AUTO_TEST_CASE(LstmLayerFloat32WithCifgWithPeepholeNoProjection,
-                     LstmLayerFloat32WithCifgWithPeepholeNoProjectionTest)
-ARMNN_AUTO_TEST_CASE(LstmLayerFloat32NoCifgNoPeepholeNoProjection,
-                     LstmLayerFloat32NoCifgNoPeepholeNoProjectionTest)
-ARMNN_AUTO_TEST_CASE(LstmLayerFloat32NoCifgWithPeepholeWithProjection,
-                     LstmLayerFloat32NoCifgWithPeepholeWithProjectionTest)
-
-// Normalization
-ARMNN_AUTO_TEST_CASE(SimpleNormalizationAcross, SimpleNormalizationAcrossTest)
-ARMNN_AUTO_TEST_CASE(SimpleNormalizationWithin, SimpleNormalizationWithinTest)
-ARMNN_AUTO_TEST_CASE(SimpleNormalizationAcrossNhwc, SimpleNormalizationAcrossNhwcTest)
-
-// ============================================================================
-// COMPARE tests
-
-ARMNN_COMPARE_REF_AUTO_TEST_CASE(CompareConv2dWithReference, CompareConvolution2dTest)
-
-ARMNN_COMPARE_REF_AUTO_TEST_CASE(CompareDepthwiseConv2dWithReferenceFloat32, CompareDepthwiseConvolution2dTest<float>)
-ARMNN_COMPARE_REF_AUTO_TEST_CASE(CompareDepthwiseConv2dWithReferenceUint8, CompareDepthwiseConvolution2dTest<uint8_t>)
-
-ARMNN_COMPARE_REF_AUTO_TEST_CASE(CompareNormalizationWithinWithReference, CompareNormalizationTest,
-                                 armnn::NormalizationAlgorithmChannel::Within,
-                                 armnn::NormalizationAlgorithmMethod::LocalBrightness)
-ARMNN_COMPARE_REF_AUTO_TEST_CASE(CompareNormalizationAcrossWithReference, CompareNormalizationTest,
-                                 armnn::NormalizationAlgorithmChannel::Across,
-                                 armnn::NormalizationAlgorithmMethod::LocalBrightness)
-
-ARMNN_COMPARE_REF_AUTO_TEST_CASE(CompareMaxPooling2dWithReference, ComparePooling2dTest, armnn::PoolingAlgorithm::Max)
-ARMNN_COMPARE_REF_AUTO_TEST_CASE(CompareMaxPooling2dWithReferenceUint8, ComparePooling2dUint8Test,
-                                 armnn::PoolingAlgorithm::Max)
-ARMNN_COMPARE_REF_AUTO_TEST_CASE(CompareAveragePooling2dWithReference, ComparePooling2dTest,
-                                 armnn::PoolingAlgorithm::Average)
-ARMNN_COMPARE_REF_AUTO_TEST_CASE(CompareAveragePooling2dWithReferenceUint8, ComparePooling2dUint8Test,
-                                 armnn::PoolingAlgorithm::Average)
-ARMNN_COMPARE_REF_AUTO_TEST_CASE(CompareL2Pooling2dWithReference, ComparePooling2dTest, armnn::PoolingAlgorithm::L2)
-ARMNN_COMPARE_REF_AUTO_TEST_CASE(UNSUPPORTED_CompareL2Pooling2dWithReferenceUint8, ComparePooling2dUint8Test,
-                                 armnn::PoolingAlgorithm::L2)
-
-ARMNN_COMPARE_REF_AUTO_TEST_CASE(CompareSoftmaxBeta1WithReference, CompareSoftmaxTest, 1.0f)
-ARMNN_COMPARE_REF_AUTO_TEST_CASE(CompareSoftmaxBeta2WithReference, CompareSoftmaxTest, 2.0f)
-
-ARMNN_COMPARE_REF_AUTO_TEST_CASE(CompareSoftmaxUint8Beta1WithReference, CompareSoftmaxUint8Test, 1.0f)
-ARMNN_COMPARE_REF_AUTO_TEST_CASE(CompareSoftmaxUint8Beta2WithReference, CompareSoftmaxUint8Test, 2.0f)
-
-ARMNN_COMPARE_REF_AUTO_TEST_CASE(CompareAddition, CompareAdditionTest)
-
-ARMNN_COMPARE_REF_AUTO_TEST_CASE(CompareMultiplicationWithReference, CompareMultiplicationTest)
-
-ARMNN_COMPARE_REF_AUTO_TEST_CASE(CompareBatchNorm, CompareBatchNormTest)
-
-ARMNN_COMPARE_REF_AUTO_TEST_CASE(ReLu1, CompareBoundedReLuTest, 1.0f, -1.0f)
-ARMNN_COMPARE_REF_AUTO_TEST_CASE(ReLu6, CompareBoundedReLuTest, 6.0f, 0.0f)
-
-// ============================================================================
-// FIXTURE tests
-
-ARMNN_COMPARE_REF_FIXTURE_TEST_CASE(CompareSigmoidActivationWithReference, ActivationFixture,
-                                    CompareActivationTest, armnn::ActivationFunction::Sigmoid, 5u)
-
-ARMNN_COMPARE_REF_FIXTURE_TEST_CASE(CompareTanhActivationWithReference, ActivationFixture,
-                                    CompareActivationTest, armnn::ActivationFunction::TanH, 5u)
-
-ARMNN_COMPARE_REF_FIXTURE_TEST_CASE(CompareLinearActivationWithReference, ActivationFixture,
-                                    CompareActivationTest, armnn::ActivationFunction::Linear, 5u)
-
-ARMNN_COMPARE_REF_FIXTURE_TEST_CASE(CompareReLuActivationWithReference, ActivationFixture,
-                                    CompareActivationTest, armnn::ActivationFunction::ReLu, 5u)
-
-ARMNN_COMPARE_REF_FIXTURE_TEST_CASE(CompareBoundedReLuActivationWithReference, ActivationFixture,
-                                    CompareActivationTest, armnn::ActivationFunction::BoundedReLu, 5u)
-ARMNN_COMPARE_REF_FIXTURE_TEST_CASE(CompareBoundedReLuActivationWithReferenceUint8, ActivationFixture,
-                                    CompareActivationUint8Test, armnn::ActivationFunction::BoundedReLu)
-
-ARMNN_COMPARE_REF_FIXTURE_TEST_CASE(CompareSoftReLuActivationWithReference, ActivationFixture,
-                                    CompareActivationTest, armnn::ActivationFunction::SoftReLu, 1u)
-
-ARMNN_COMPARE_REF_FIXTURE_TEST_CASE(CompareLeakyReLuActivationWithReference, ActivationFixture,
-                                    CompareActivationTest, armnn::ActivationFunction::LeakyReLu, 5u)
-
-ARMNN_COMPARE_REF_FIXTURE_TEST_CASE(CompareAbsActivationWithReference, ActivationFixture,
-                                    CompareActivationTest, armnn::ActivationFunction::Abs, 5u)
-
-ARMNN_COMPARE_REF_FIXTURE_TEST_CASE(CompareSqrtActivationWithReference, PositiveActivationFixture,
-                                    CompareActivationTest, armnn::ActivationFunction::Sqrt, 5u)
-
-ARMNN_COMPARE_REF_FIXTURE_TEST_CASE(CompareSquareActivationWithReference, ActivationFixture,
-                                    CompareActivationTest, armnn::ActivationFunction::Square, 5u)
-BOOST_AUTO_TEST_SUITE_END()
diff --git a/src/backends/test/ClContextControlFixture.hpp b/src/backends/test/ClContextControlFixture.hpp
deleted file mode 100644
index fd53e3fcf3..0000000000
--- a/src/backends/test/ClContextControlFixture.hpp
+++ /dev/null
@@ -1,34 +0,0 @@
-//
-// Copyright © 2017 Arm Ltd. All rights reserved.
-// SPDX-License-Identifier: MIT
-//
-
-#pragma once
-
-#include <backends/cl/ClContextControl.hpp>
-
-template<bool ProfilingEnabled>
-struct ClContextControlFixtureBase
-{
-    static ClContextControlFixtureBase*& Instance()
-    {
-        static ClContextControlFixtureBase* s_Instance = nullptr;
-        return s_Instance;
-    }
-
-    // Initialising ClContextControl to ensure OpenCL is loaded correctly for each test case
-    ClContextControlFixtureBase()
-        : m_ClContextControl(nullptr, ProfilingEnabled)
-    {
-        Instance() = this;
-    }
-    ~ClContextControlFixtureBase()
-    {
-        Instance() = nullptr;
-    }
-
-    armnn::ClContextControl m_ClContextControl;
-};
-
-using ClContextControlFixture = ClContextControlFixtureBase<false>;
-using ClProfilingContextControlFixture = ClContextControlFixtureBase<true>;
diff --git a/src/backends/test/CreateWorkloadCl.cpp b/src/backends/test/CreateWorkloadCl.cpp
deleted file mode 100644
index d56bad2bb9..0000000000
--- a/src/backends/test/CreateWorkloadCl.cpp
+++ /dev/null
@@ -1,676 +0,0 @@
-//
-// Copyright © 2017 Arm Ltd. All rights reserved.
-// SPDX-License-Identifier: MIT
-//
-#include <backends/cl/ClWorkloadFactory.hpp>
-#include <backends/reference/RefWorkloadFactory.hpp>
-#include <backends/MemCopyWorkload.hpp>
-#include <backends/cl/workloads/ClWorkloadUtils.hpp>
-#include <backends/cl/workloads/ClWorkloads.hpp>
-#include <backends/cl/ClTensorHandle.hpp>
-#include "ClContextControlFixture.hpp"
-
-#include <test/CreateWorkloadClNeon.hpp>
-
-boost::test_tools::predicate_result CompareIClTensorHandleShape(IClTensorHandle*                    tensorHandle,
-                                                                std::initializer_list<unsigned int> expectedDimensions)
-{
-    return CompareTensorHandleShape<IClTensorHandle>(tensorHandle, expectedDimensions);
-}
-
-BOOST_FIXTURE_TEST_SUITE(CreateWorkloadCl, ClContextControlFixture)
-
-template <typename ActivationWorkloadType, armnn::DataType DataType>
-static void ClCreateActivationWorkloadTest()
-{
-    Graph graph;
-    ClWorkloadFactory factory;
-
-    auto workload = CreateActivationWorkloadTest<ActivationWorkloadType, DataType>(factory, graph);
-
-    // Checks that inputs/outputs are as we expect them (see definition of CreateActivationWorkloadTest).
-    ActivationQueueDescriptor queueDescriptor = workload->GetData();
-    auto inputHandle = boost::polymorphic_downcast<IClTensorHandle*>(queueDescriptor.m_Inputs[0]);
-    auto outputHandle = boost::polymorphic_downcast<IClTensorHandle*>(queueDescriptor.m_Outputs[0]);
-
-    BOOST_TEST(CompareIClTensorHandleShape(inputHandle, {1}));
-    BOOST_TEST(CompareIClTensorHandleShape(outputHandle, {1}));
-}
-
-BOOST_AUTO_TEST_CASE(CreateActivationFloatWorkload)
-{
-    ClCreateActivationWorkloadTest<ClActivationFloatWorkload, armnn::DataType::Float32>();
-}
-
-BOOST_AUTO_TEST_CASE(CreateActivationFloat16Workload)
-{
-    ClCreateActivationWorkloadTest<ClActivationFloatWorkload, armnn::DataType::Float16>();
-}
-
-template <typename WorkloadType,
-          typename DescriptorType,
-          typename LayerType,
-          armnn::DataType DataType>
-static void ClCreateArithmethicWorkloadTest()
-{
-    Graph graph;
-    ClWorkloadFactory factory;
-    auto workload = CreateArithmeticWorkloadTest<WorkloadType, DescriptorType, LayerType, DataType>(factory, graph);
-
-    // Checks that inputs/outputs are as we expect them (see definition of CreateArithmeticWorkloadTest).
-    DescriptorType queueDescriptor = workload->GetData();
-    auto inputHandle1 = boost::polymorphic_downcast<IClTensorHandle*>(queueDescriptor.m_Inputs[0]);
-    auto inputHandle2 = boost::polymorphic_downcast<IClTensorHandle*>(queueDescriptor.m_Inputs[1]);
-    auto outputHandle = boost::polymorphic_downcast<IClTensorHandle*>(queueDescriptor.m_Outputs[0]);
-    BOOST_TEST(CompareIClTensorHandleShape(inputHandle1, {2, 3}));
-    BOOST_TEST(CompareIClTensorHandleShape(inputHandle2, {2, 3}));
-    BOOST_TEST(CompareIClTensorHandleShape(outputHandle, {2, 3}));
-}
-
-BOOST_AUTO_TEST_CASE(CreateAdditionFloatWorkload)
-{
-    ClCreateArithmethicWorkloadTest<ClAdditionWorkload,
-                                    AdditionQueueDescriptor,
-                                    AdditionLayer,
-                                    armnn::DataType::Float32>();
-}
-
-BOOST_AUTO_TEST_CASE(CreateAdditionFloat16Workload)
-{
-    ClCreateArithmethicWorkloadTest<ClAdditionWorkload,
-                                    AdditionQueueDescriptor,
-                                    AdditionLayer,
-                                    armnn::DataType::Float16>();
-}
-
-BOOST_AUTO_TEST_CASE(CreateSubtractionFloatWorkload)
-{
-    ClCreateArithmethicWorkloadTest<ClSubtractionWorkload,
-                                    SubtractionQueueDescriptor,
-                                    SubtractionLayer,
-                                    armnn::DataType::Float32>();
-}
-
-BOOST_AUTO_TEST_CASE(CreateSubtractionFloat16Workload)
-{
-    ClCreateArithmethicWorkloadTest<ClSubtractionWorkload,
-                                    SubtractionQueueDescriptor,
-                                    SubtractionLayer,
-                                    armnn::DataType::Float16>();
-}
-
-BOOST_AUTO_TEST_CASE(CreateMultiplicationFloatWorkloadTest)
-{
-    ClCreateArithmethicWorkloadTest<ClMultiplicationWorkload,
-                                    MultiplicationQueueDescriptor,
-                                    MultiplicationLayer,
-                                    armnn::DataType::Float32>();
-}
-
-BOOST_AUTO_TEST_CASE(CreateMultiplicationFloat16WorkloadTest)
-{
-    ClCreateArithmethicWorkloadTest<ClMultiplicationWorkload,
-                                    MultiplicationQueueDescriptor,
-                                    MultiplicationLayer,
-                                    armnn::DataType::Float16>();
-}
-
-BOOST_AUTO_TEST_CASE(CreateMultiplicationUint8WorkloadTest)
-{
-    ClCreateArithmethicWorkloadTest<ClMultiplicationWorkload,
-                                    MultiplicationQueueDescriptor,
-                                    MultiplicationLayer,
-                                    armnn::DataType::QuantisedAsymm8>();
-}
-
-BOOST_AUTO_TEST_CASE(CreateDivisionFloatWorkloadTest)
-{
-    ClCreateArithmethicWorkloadTest<ClDivisionFloatWorkload,
-                                    DivisionQueueDescriptor,
-                                    DivisionLayer,
-                                    armnn::DataType::Float32>();
-}
-
-BOOST_AUTO_TEST_CASE(CreateDivisionFloat16WorkloadTest)
-{
-    ClCreateArithmethicWorkloadTest<ClDivisionFloatWorkload,
-                                    DivisionQueueDescriptor,
-                                    DivisionLayer,
-                                    armnn::DataType::Float16>();
-}
-
-template <typename BatchNormalizationWorkloadType, armnn::DataType DataType>
-static void ClCreateBatchNormalizationWorkloadTest()
-{
-    Graph graph;
-    ClWorkloadFactory factory;
-
-    auto workload = CreateBatchNormalizationWorkloadTest<BatchNormalizationWorkloadType, DataType>
-                    (factory, graph);
-
-    // Checks that inputs/outputs are as we expect them (see definition of CreateBatchNormalizationWorkloadTest).
-    BatchNormalizationQueueDescriptor queueDescriptor = workload->GetData();
-    auto inputHandle = boost::polymorphic_downcast<IClTensorHandle*>(queueDescriptor.m_Inputs[0]);
-    auto outputHandle = boost::polymorphic_downcast<IClTensorHandle*>(queueDescriptor.m_Outputs[0]);
-
-    BOOST_TEST(CompareIClTensorHandleShape(inputHandle, {2, 3, 1, 1}));
-    BOOST_TEST(CompareIClTensorHandleShape(outputHandle, {2, 3, 1, 1}));
-}
-
-BOOST_AUTO_TEST_CASE(CreateBatchNormalizationFloatWorkload)
-{
-    ClCreateBatchNormalizationWorkloadTest<ClBatchNormalizationFloatWorkload, armnn::DataType::Float32>();
-}
-
-BOOST_AUTO_TEST_CASE(CreateBatchNormalizationFloat16Workload)
-{
-    ClCreateBatchNormalizationWorkloadTest<ClBatchNormalizationFloatWorkload, armnn::DataType::Float16>();
-}
-
-BOOST_AUTO_TEST_CASE(CreateConvertFp16ToFp32Workload)
-{
-    Graph graph;
-    ClWorkloadFactory factory;
-    auto workload = CreateConvertFp16ToFp32WorkloadTest<ClConvertFp16ToFp32Workload>(factory, graph);
-
-    ConvertFp16ToFp32QueueDescriptor queueDescriptor = workload->GetData();
-    auto inputHandle  = boost::polymorphic_downcast<IClTensorHandle*>(queueDescriptor.m_Inputs[0]);
-    auto outputHandle = boost::polymorphic_downcast<IClTensorHandle*>(queueDescriptor.m_Outputs[0]);
-
-    BOOST_TEST(CompareIClTensorHandleShape(inputHandle, {3, 2, 3}));
-    BOOST_TEST(CompareIClTensorHandleShape(outputHandle, {3, 2, 3}));
-    BOOST_TEST((inputHandle->GetTensor().info()->data_type() == arm_compute::DataType::F16));
-    BOOST_TEST((outputHandle->GetTensor().info()->data_type() == arm_compute::DataType::F32));
-}
-
-BOOST_AUTO_TEST_CASE(CreateConvertFp32ToFp16Workload)
-{
-    Graph graph;
-    ClWorkloadFactory factory;
-    auto workload = CreateConvertFp32ToFp16WorkloadTest<ClConvertFp32ToFp16Workload>(factory, graph);
-
-    ConvertFp32ToFp16QueueDescriptor queueDescriptor = workload->GetData();
-    auto inputHandle  = boost::polymorphic_downcast<IClTensorHandle*>(queueDescriptor.m_Inputs[0]);
-    auto outputHandle = boost::polymorphic_downcast<IClTensorHandle*>(queueDescriptor.m_Outputs[0]);
-
-    BOOST_TEST(CompareIClTensorHandleShape(inputHandle, {3, 2, 3}));
-    BOOST_TEST(CompareIClTensorHandleShape(outputHandle, {3, 2, 3}));
-    BOOST_TEST((inputHandle->GetTensor().info()->data_type() == arm_compute::DataType::F32));
-    BOOST_TEST((outputHandle->GetTensor().info()->data_type() == arm_compute::DataType::F16));
-}
-
-template <typename Convolution2dWorkloadType, typename armnn::DataType DataType>
-static void ClConvolution2dWorkloadTest(DataLayout dataLayout)
-{
-    Graph graph;
-    ClWorkloadFactory factory;
-    auto workload = CreateConvolution2dWorkloadTest<ClConvolution2dWorkload, DataType>(factory,
-                                                                                       graph,
-                                                                                       dataLayout);
-
-    std::initializer_list<unsigned int> inputShape  = (dataLayout == DataLayout::NCHW) ?
-            std::initializer_list<unsigned int>({2, 3, 8, 16}) : std::initializer_list<unsigned int>({2, 8, 16, 3});
-    std::initializer_list<unsigned int> outputShape = (dataLayout == DataLayout::NCHW) ?
-            std::initializer_list<unsigned int>({2, 2, 2, 10}) : std::initializer_list<unsigned int>({2, 2, 10, 2});
-
-    // Checks that outputs and inputs are as we expect them (see definition of CreateConvolution2dWorkloadTest).
-    Convolution2dQueueDescriptor queueDescriptor = workload->GetData();
-    auto inputHandle  = boost::polymorphic_downcast<IClTensorHandle*>(queueDescriptor.m_Inputs[0]);
-    auto outputHandle = boost::polymorphic_downcast<IClTensorHandle*>(queueDescriptor.m_Outputs[0]);
-    BOOST_TEST(CompareIClTensorHandleShape(inputHandle, inputShape));
-    BOOST_TEST(CompareIClTensorHandleShape(outputHandle, outputShape));
-}
-
-BOOST_AUTO_TEST_CASE(CreateConvolution2dFloatNchwWorkload)
-{
-    ClConvolution2dWorkloadTest<ClConvolution2dWorkload, armnn::DataType::Float32>(DataLayout::NCHW);
-}
-
-BOOST_AUTO_TEST_CASE(CreateConvolution2dFloatNhwcWorkload)
-{
-    ClConvolution2dWorkloadTest<ClConvolution2dWorkload, armnn::DataType::Float32>(DataLayout::NHWC);
-}
-
-BOOST_AUTO_TEST_CASE(CreateConvolution2dFloat16NchwWorkload)
-{
-    ClConvolution2dWorkloadTest<ClConvolution2dWorkload, armnn::DataType::Float16>(DataLayout::NCHW);
-}
-
-BOOST_AUTO_TEST_CASE(CreateConvolution2dFloat16NhwcWorkload)
-{
-    ClConvolution2dWorkloadTest<ClConvolution2dWorkload, armnn::DataType::Float16>(DataLayout::NHWC);
-}
-
-template <typename Convolution2dWorkloadType, typename armnn::DataType DataType>
-static void ClDirectConvolution2dWorkloadTest()
-{
-    Graph graph;
-    ClWorkloadFactory factory;
-    auto workload = CreateDirectConvolution2dWorkloadTest<ClConvolution2dWorkload, DataType>(factory, graph);
-
-    // Checks that outputs and inputs are as we expect them (see definition of CreateDirectConvolution2dWorkloadTest).
-    Convolution2dQueueDescriptor queueDescriptor = workload->GetData();
-    auto inputHandle  = boost::polymorphic_downcast<IClTensorHandle*>(queueDescriptor.m_Inputs[0]);
-    auto outputHandle = boost::polymorphic_downcast<IClTensorHandle*>(queueDescriptor.m_Outputs[0]);
-    BOOST_TEST(CompareIClTensorHandleShape(inputHandle, {2, 3, 6, 6}));
-    BOOST_TEST(CompareIClTensorHandleShape(outputHandle, {2, 2, 6, 6}));
-}
-
-BOOST_AUTO_TEST_CASE(CreateDirectConvolution2dFloatWorkload)
-{
-    ClDirectConvolution2dWorkloadTest<ClConvolution2dWorkload, armnn::DataType::Float32>();
-}
-
-BOOST_AUTO_TEST_CASE(CreateDirectConvolution2dFloat16Workload)
-{
-    ClDirectConvolution2dWorkloadTest<ClConvolution2dWorkload, armnn::DataType::Float16>();
-}
-
-BOOST_AUTO_TEST_CASE(CreateDirectConvolution2dUint8Workload)
-{
-    ClDirectConvolution2dWorkloadTest<ClConvolution2dWorkload, armnn::DataType::QuantisedAsymm8>();
-}
-
-template <typename FullyConnectedWorkloadType, typename armnn::DataType DataType>
-static void ClCreateFullyConnectedWorkloadTest()
-{
-    Graph graph;
-    ClWorkloadFactory factory;
-    auto workload =
-        CreateFullyConnectedWorkloadTest<FullyConnectedWorkloadType, DataType>(factory, graph);
-
-    // Checks that outputs and inputs are as we expect them (see definition of CreateFullyConnectedWorkloadTest).
-    FullyConnectedQueueDescriptor queueDescriptor = workload->GetData();
-    auto inputHandle = boost::polymorphic_downcast<IClTensorHandle*>(queueDescriptor.m_Inputs[0]);
-    auto outputHandle = boost::polymorphic_downcast<IClTensorHandle*>(queueDescriptor.m_Outputs[0]);
-    BOOST_TEST(CompareIClTensorHandleShape(inputHandle, {3, 1, 4, 5}));
-    BOOST_TEST(CompareIClTensorHandleShape(outputHandle, {3, 7}));
-}
-
-
-BOOST_AUTO_TEST_CASE(CreateFullyConnectedFloatWorkloadTest)
-{
-    ClCreateFullyConnectedWorkloadTest<ClFullyConnectedWorkload, armnn::DataType::Float32>();
-}
-
-BOOST_AUTO_TEST_CASE(CreateFullyConnectedFloat16WorkloadTest)
-{
-    ClCreateFullyConnectedWorkloadTest<ClFullyConnectedWorkload, armnn::DataType::Float16>();
-}
-
-template <typename NormalizationWorkloadType, typename armnn::DataType DataType>
-static void ClNormalizationWorkloadTest(DataLayout dataLayout)
-{
-    Graph graph;
-    ClWorkloadFactory factory;
-
-    auto workload = CreateNormalizationWorkloadTest<NormalizationWorkloadType, DataType>
-                    (factory, graph, dataLayout);
-
-    // Checks that inputs/outputs are as we expect them (see definition of CreateNormalizationWorkloadTest).
-    NormalizationQueueDescriptor queueDescriptor = workload->GetData();
-    auto inputHandle  = boost::polymorphic_downcast<IClTensorHandle*>(queueDescriptor.m_Inputs[0]);
-    auto outputHandle = boost::polymorphic_downcast<IClTensorHandle*>(queueDescriptor.m_Outputs[0]);
-
-    BOOST_TEST(CompareIClTensorHandleShape(inputHandle, {3, 5, 5, 1}));
-    BOOST_TEST(CompareIClTensorHandleShape(outputHandle, {3, 5, 5, 1}));
-}
-
-BOOST_AUTO_TEST_CASE(CreateNormalizationFloat32NchwWorkload)
-{
-    ClNormalizationWorkloadTest<ClNormalizationFloatWorkload, armnn::DataType::Float32>(DataLayout::NCHW);
-}
-
-BOOST_AUTO_TEST_CASE(CreateNormalizationFloat16NchwWorkload)
-{
-    ClNormalizationWorkloadTest<ClNormalizationFloatWorkload, armnn::DataType::Float16>(DataLayout::NCHW);
-}
-
-BOOST_AUTO_TEST_CASE(CreateNormalizationFloat32NhwcWorkload)
-{
-    ClNormalizationWorkloadTest<ClNormalizationFloatWorkload, armnn::DataType::Float32>(DataLayout::NHWC);
-}
-
-BOOST_AUTO_TEST_CASE(CreateNormalizationFloat16NhwcWorkload)
-{
-    ClNormalizationWorkloadTest<ClNormalizationFloatWorkload, armnn::DataType::Float16>(DataLayout::NHWC);
-}
-
-template <typename Pooling2dWorkloadType, typename armnn::DataType DataType>
-static void ClPooling2dWorkloadTest(DataLayout dataLayout)
-{
-    Graph graph;
-    ClWorkloadFactory factory;
-
-    auto workload = CreatePooling2dWorkloadTest<Pooling2dWorkloadType, DataType>(factory, graph, dataLayout);
-
-    std::initializer_list<unsigned int> inputShape  = (dataLayout == DataLayout::NCHW) ?
-            std::initializer_list<unsigned int>({3, 2, 5, 5}) : std::initializer_list<unsigned int>({3, 5, 5, 2});
-    std::initializer_list<unsigned int> outputShape = (dataLayout == DataLayout::NCHW) ?
-            std::initializer_list<unsigned int>({3, 2, 2, 4}) : std::initializer_list<unsigned int>({3, 2, 4, 2});
-
-    // Check that inputs/outputs are as we expect them (see definition of CreatePooling2dWorkloadTest).
-    Pooling2dQueueDescriptor queueDescriptor = workload->GetData();
-    auto inputHandle  = boost::polymorphic_downcast<IClTensorHandle*>(queueDescriptor.m_Inputs[0]);
-    auto outputHandle = boost::polymorphic_downcast<IClTensorHandle*>(queueDescriptor.m_Outputs[0]);
-
-    BOOST_TEST(CompareIClTensorHandleShape(inputHandle, inputShape));
-    BOOST_TEST(CompareIClTensorHandleShape(outputHandle, outputShape));
-}
-
-BOOST_AUTO_TEST_CASE(CreatePooling2dFloatNchwWorkload)
-{
-    ClPooling2dWorkloadTest<ClPooling2dFloatWorkload, armnn::DataType::Float32>(DataLayout::NCHW);
-}
-
-BOOST_AUTO_TEST_CASE(CreatePooling2dFloatNhwcWorkload)
-{
-    ClPooling2dWorkloadTest<ClPooling2dFloatWorkload, armnn::DataType::Float32>(DataLayout::NHWC);
-}
-
-BOOST_AUTO_TEST_CASE(CreatePooling2dFloat16NchwWorkload)
-{
-    ClPooling2dWorkloadTest<ClPooling2dFloatWorkload, armnn::DataType::Float16>(DataLayout::NCHW);
-}
-
-BOOST_AUTO_TEST_CASE(CreatePooling2dFloat16NhwcWorkload)
-{
-    ClPooling2dWorkloadTest<ClPooling2dFloatWorkload, armnn::DataType::Float16>(DataLayout::NHWC);
-}
-
-template <typename ReshapeWorkloadType, typename armnn::DataType DataType>
-static void ClCreateReshapeWorkloadTest()
-{
-    Graph graph;
-    ClWorkloadFactory factory;
-
-    auto workload = CreateReshapeWorkloadTest<ReshapeWorkloadType, DataType>(factory, graph);
-
-    // Checks that outputs and inputs are as we expect them (see definition of CreateReshapeWorkloadTest).
-    ReshapeQueueDescriptor queueDescriptor = workload->GetData();
-    auto inputHandle = boost::polymorphic_downcast<IClTensorHandle*>(queueDescriptor.m_Inputs[0]);
-    auto outputHandle = boost::polymorphic_downcast<IClTensorHandle*>(queueDescriptor.m_Outputs[0]);
-
-    BOOST_TEST(CompareIClTensorHandleShape(inputHandle, {4, 1}));
-    BOOST_TEST(CompareIClTensorHandleShape(outputHandle, {4})); // Leading size 1 dimensions are collapsed by ACL.
-}
-
-BOOST_AUTO_TEST_CASE(CreateReshapeFloatWorkload)
-{
-    ClCreateReshapeWorkloadTest<ClReshapeFloatWorkload, armnn::DataType::Float32>();
-}
-
-BOOST_AUTO_TEST_CASE(CreateReshapeFloat16Workload)
-{
-    ClCreateReshapeWorkloadTest<ClReshapeFloatWorkload, armnn::DataType::Float16>();
-}
-
-BOOST_AUTO_TEST_CASE(CreateReshapeUint8Workload)
-{
-    ClCreateReshapeWorkloadTest<ClReshapeUint8Workload, armnn::DataType::QuantisedAsymm8>();
-}
-
-template <typename SoftmaxWorkloadType, typename armnn::DataType DataType>
-static void ClSoftmaxWorkloadTest()
-{
-    Graph graph;
-    ClWorkloadFactory factory;
-
-    auto workload = CreateSoftmaxWorkloadTest<SoftmaxWorkloadType, DataType>(factory, graph);
-
-    // Checks that inputs/outputs are as we expect them (see definition of ClSoftmaxFloatWorkload).
-    SoftmaxQueueDescriptor queueDescriptor = workload->GetData();
-    auto inputHandle = boost::polymorphic_downcast<IClTensorHandle*>(queueDescriptor.m_Inputs[0]);
-    auto outputHandle = boost::polymorphic_downcast<IClTensorHandle*>(queueDescriptor.m_Outputs[0]);
-
-    BOOST_TEST(CompareIClTensorHandleShape(inputHandle, {4, 1}));
-    BOOST_TEST(CompareIClTensorHandleShape(outputHandle, {4, 1}));
-}
-
-
-BOOST_AUTO_TEST_CASE(CreateSoftmaxFloatWorkloadTest)
-{
-    ClSoftmaxWorkloadTest<ClSoftmaxFloatWorkload, armnn::DataType::Float32>();
-}
-
-BOOST_AUTO_TEST_CASE(CreateSoftmaxFloat16WorkloadTest)
-{
-    ClSoftmaxWorkloadTest<ClSoftmaxFloatWorkload, armnn::DataType::Float16>();
-}
-
-template <typename armnn::DataType DataType>
-static void ClSplitterWorkloadTest()
-{
-    Graph graph;
-    ClWorkloadFactory factory;
-
-    auto workload = CreateSplitterWorkloadTest<ClSplitterWorkload, DataType>(factory, graph);
-
-    // Checks that outputs are as we expect them (see definition of CreateSplitterWorkloadTest).
-    SplitterQueueDescriptor queueDescriptor = workload->GetData();
-    auto inputHandle = boost::polymorphic_downcast<IClTensorHandle*>(queueDescriptor.m_Inputs[0]);
-    BOOST_TEST(CompareIClTensorHandleShape(inputHandle, {5, 7, 7}));
-
-    auto outputHandle1 = boost::polymorphic_downcast<IClTensorHandle*>(queueDescriptor.m_Outputs[1]);
-    BOOST_TEST(CompareIClTensorHandleShape(outputHandle1, {2, 7, 7}));
-
-    auto outputHandle2 = boost::polymorphic_downcast<IClTensorHandle*>(queueDescriptor.m_Outputs[2]);
-    BOOST_TEST(CompareIClTensorHandleShape(outputHandle2, {2, 7, 7}));
-
-    auto outputHandle0 = boost::polymorphic_downcast<IClTensorHandle*>(queueDescriptor.m_Outputs[0]);
-    // NOTE: At the moment the CL collapses the tensor to a 2 dim when dimension zero = 1
-    //       we are raising this difference between the NEON and CL libs as an issue with the compute library team.
-    BOOST_TEST(CompareIClTensorHandleShape(outputHandle0, {7, 7}));
-}
-
-BOOST_AUTO_TEST_CASE(CreateSplitterFloatWorkload)
-{
-    ClSplitterWorkloadTest<armnn::DataType::Float32>();
-}
-
-BOOST_AUTO_TEST_CASE(CreateSplitterFloat16Workload)
-{
-    ClSplitterWorkloadTest<armnn::DataType::Float16>();
-}
-
-template <typename MergerWorkloadType, typename armnn::DataType DataType>
-static void ClSplitterMergerTest()
-{
-    // Tests that it is possible to decide which output of the splitter layer
-    // should be lined to which input of the merger layer.
-    // We test that is is possible to specify 0th output
-    // of the splitter to be the 1st input to the merger and the 1st output of the splitter  to be 0th input
-    // of the merger.
-
-    Graph graph;
-    ClWorkloadFactory factory;
-
-    auto workloads =
-        CreateSplitterMergerWorkloadTest<ClSplitterWorkload, MergerWorkloadType, DataType>
-            (factory, graph);
-
-    auto wlSplitter = std::move(workloads.first);
-    auto wlMerger = std::move(workloads.second);
-
-    //Checks that the index of inputs/outputs matches what we declared on InputDescriptor construction.
-    armnn::ClSubTensorHandle* sOut0 = dynamic_cast<armnn::ClSubTensorHandle*>(wlSplitter->GetData().m_Outputs[0]);
-    armnn::ClSubTensorHandle* sOut1 = dynamic_cast<armnn::ClSubTensorHandle*>(wlSplitter->GetData().m_Outputs[1]);
-    armnn::ClSubTensorHandle* mIn0 = dynamic_cast<armnn::ClSubTensorHandle*>(wlMerger->GetData().m_Inputs[0]);
-    armnn::ClSubTensorHandle* mIn1 = dynamic_cast<armnn::ClSubTensorHandle*>(wlMerger->GetData().m_Inputs[1]);
-
-    BOOST_TEST(sOut0);
-    BOOST_TEST(sOut1);
-    BOOST_TEST(mIn0);
-    BOOST_TEST(mIn1);
-
-    //Fliped order of inputs/outputs.
-    bool validDataPointers = (sOut0 == mIn1) && (sOut1 == mIn0);
-    BOOST_TEST(validDataPointers);
-
-
-    //Also make sure that the inputs are subtensors of one tensor and outputs are sub tensors of another tensor.
-    bool validSubTensorParents = (mIn0->GetTensor().parent() == mIn1->GetTensor().parent())
-                                    && (sOut0->GetTensor().parent() == sOut1->GetTensor().parent());
-
-    BOOST_TEST(validSubTensorParents);
-}
-
-BOOST_AUTO_TEST_CASE(CreateSplitterMergerFloatWorkload)
-{
-    ClSplitterMergerTest<ClMergerFloatWorkload, armnn::DataType::Float32>();
-}
-
-BOOST_AUTO_TEST_CASE(CreateSplitterMergerFloat16Workload)
-{
-    ClSplitterMergerTest<ClMergerFloatWorkload, armnn::DataType::Float16>();
-}
-
-
-BOOST_AUTO_TEST_CASE(CreateSingleOutputMultipleInputs)
-{
-    // Test that it is possible to assign multiple (two) different layers to each of the outputs of a splitter layer.
-    // We create a splitter with two outputs. That each of those outputs is used by two different activation layers.
-
-    Graph graph;
-    ClWorkloadFactory factory;
-    std::unique_ptr<ClSplitterWorkload> wlSplitter;
-    std::unique_ptr<ClActivationFloatWorkload> wlActiv0_0;
-    std::unique_ptr<ClActivationFloatWorkload> wlActiv0_1;
-    std::unique_ptr<ClActivationFloatWorkload> wlActiv1_0;
-    std::unique_ptr<ClActivationFloatWorkload> wlActiv1_1;
-
-    CreateSplitterMultipleInputsOneOutputWorkloadTest<ClSplitterWorkload,
-        ClActivationFloatWorkload, armnn::DataType::Float32>(factory, graph, wlSplitter, wlActiv0_0, wlActiv0_1,
-                                                               wlActiv1_0, wlActiv1_1);
-
-    //Checks that the index of inputs/outputs matches what we declared on InputDescriptor construction.
-    armnn::ClSubTensorHandle* sOut0 = dynamic_cast<armnn::ClSubTensorHandle*>(wlSplitter->GetData().m_Outputs[0]);
-    armnn::ClSubTensorHandle* sOut1 = dynamic_cast<armnn::ClSubTensorHandle*>(wlSplitter->GetData().m_Outputs[1]);
-    armnn::ClSubTensorHandle* activ0_0Im = dynamic_cast<armnn::ClSubTensorHandle*>(wlActiv0_0->GetData().m_Inputs[0]);
-    armnn::ClSubTensorHandle* activ0_1Im = dynamic_cast<armnn::ClSubTensorHandle*>(wlActiv0_1->GetData().m_Inputs[0]);
-    armnn::ClSubTensorHandle* activ1_0Im = dynamic_cast<armnn::ClSubTensorHandle*>(wlActiv1_0->GetData().m_Inputs[0]);
-    armnn::ClSubTensorHandle* activ1_1Im = dynamic_cast<armnn::ClSubTensorHandle*>(wlActiv1_1->GetData().m_Inputs[0]);
-
-
-    BOOST_TEST(sOut0);
-    BOOST_TEST(sOut1);
-    BOOST_TEST(activ0_0Im);
-    BOOST_TEST(activ0_1Im);
-    BOOST_TEST(activ1_0Im);
-    BOOST_TEST(activ1_1Im);
-
-    bool validDataPointers = (sOut0 == activ0_0Im) && (sOut0 == activ0_1Im) &&
-                             (sOut1 == activ1_0Im) && (sOut1 == activ1_1Im);
-
-    BOOST_TEST(validDataPointers);
-}
-
-BOOST_AUTO_TEST_CASE(CreateMemCopyWorkloadsCl)
-{
-    ClWorkloadFactory    factory;
-    CreateMemCopyWorkloads<IClTensorHandle>(factory);
-}
-
-template <typename L2NormalizationWorkloadType, typename armnn::DataType DataType>
-static void ClL2NormalizationWorkloadTest(DataLayout dataLayout)
-{
-    Graph graph;
-    ClWorkloadFactory factory;
-
-    auto workload = CreateL2NormalizationWorkloadTest<L2NormalizationWorkloadType, DataType>
-        (factory, graph, dataLayout);
-
-    // Checks that inputs/outputs are as we expect them (see definition of CreateNormalizationWorkloadTest).
-    L2NormalizationQueueDescriptor queueDescriptor = workload->GetData();
-    auto inputHandle = boost::polymorphic_downcast<IClTensorHandle*>(queueDescriptor.m_Inputs[0]);
-    auto outputHandle = boost::polymorphic_downcast<IClTensorHandle*>(queueDescriptor.m_Outputs[0]);
-
-    BOOST_TEST(CompareIClTensorHandleShape(inputHandle, { 5, 20, 50, 67 }));
-    BOOST_TEST(CompareIClTensorHandleShape(outputHandle, { 5, 20, 50, 67 }));
-}
-
-BOOST_AUTO_TEST_CASE(CreateL2NormalizationFloatNchwWorkload)
-{
-    ClL2NormalizationWorkloadTest<ClL2NormalizationFloatWorkload, armnn::DataType::Float32>(DataLayout::NCHW);
-}
-
-BOOST_AUTO_TEST_CASE(CreateL2NormalizationFloatNhwcWorkload)
-{
-    ClL2NormalizationWorkloadTest<ClL2NormalizationFloatWorkload, armnn::DataType::Float32>(DataLayout::NHWC);
-}
-
-BOOST_AUTO_TEST_CASE(CreateL2NormalizationFloat16NchwWorkload)
-{
-    ClL2NormalizationWorkloadTest<ClL2NormalizationFloatWorkload, armnn::DataType::Float16>(DataLayout::NCHW);
-}
-
-BOOST_AUTO_TEST_CASE(CreateL2NormalizationFloat16NhwcWorkload)
-{
-    ClL2NormalizationWorkloadTest<ClL2NormalizationFloatWorkload, armnn::DataType::Float16>(DataLayout::NHWC);
-}
-
-template <typename LstmWorkloadType>
-static void ClCreateLstmWorkloadTest()
-{
-    Graph graph;
-    ClWorkloadFactory factory;
-    auto workload = CreateLstmWorkloadTest<LstmWorkloadType>(factory, graph);
-
-    LstmQueueDescriptor queueDescriptor = workload->GetData();
-    auto inputHandle  = boost::polymorphic_downcast<IClTensorHandle*>(queueDescriptor.m_Inputs[0]);
-    auto outputHandle = boost::polymorphic_downcast<IClTensorHandle*>(queueDescriptor.m_Outputs[1]);
-    BOOST_TEST(CompareIClTensorHandleShape(inputHandle, { 2, 2 }));
-    BOOST_TEST(CompareIClTensorHandleShape(outputHandle, { 2, 4 }));
-}
-
-BOOST_AUTO_TEST_CASE(CreateLSTMWorkloadFloatWorkload)
-{
-    ClCreateLstmWorkloadTest<ClLstmFloatWorkload>();
-}
-
-template <typename ResizeBilinearWorkloadType, typename armnn::DataType DataType>
-static void ClResizeBilinearWorkloadTest(DataLayout dataLayout)
-{
-    Graph graph;
-    ClWorkloadFactory factory;
-
-    auto workload = CreateResizeBilinearWorkloadTest<ResizeBilinearWorkloadType, DataType>(factory, graph, dataLayout);
-
-    // Checks that inputs/outputs are as we expect them (see definition of CreateResizeBilinearWorkloadTest).
-    ResizeBilinearQueueDescriptor queueDescriptor = workload->GetData();
-    auto inputHandle = boost::polymorphic_downcast<IClTensorHandle*>(queueDescriptor.m_Inputs[0]);
-    auto outputHandle = boost::polymorphic_downcast<IClTensorHandle*>(queueDescriptor.m_Outputs[0]);
-
-    switch (dataLayout)
-    {
-        case DataLayout::NHWC:
-            BOOST_TEST(CompareIClTensorHandleShape(inputHandle, { 2, 4, 4, 3 }));
-            BOOST_TEST(CompareIClTensorHandleShape(outputHandle, { 2, 2, 2, 3 }));
-            break;
-        default: // NCHW
-            BOOST_TEST(CompareIClTensorHandleShape(inputHandle, { 2, 3, 4, 4 }));
-            BOOST_TEST(CompareIClTensorHandleShape(outputHandle, { 2, 3, 2, 2 }));
-    }
-}
-
-BOOST_AUTO_TEST_CASE(CreateResizeBilinearFloat32NchwWorkload)
-{
-    ClResizeBilinearWorkloadTest<ClResizeBilinearFloatWorkload, armnn::DataType::Float32>(DataLayout::NCHW);
-}
-
-BOOST_AUTO_TEST_CASE(CreateResizeBilinearFloat16NchwWorkload)
-{
-    ClResizeBilinearWorkloadTest<ClResizeBilinearFloatWorkload, armnn::DataType::Float16>(DataLayout::NCHW);
-}
-
-BOOST_AUTO_TEST_CASE(CreateResizeBilinearFloat32NhwcWorkload)
-{
-    ClResizeBilinearWorkloadTest<ClResizeBilinearFloatWorkload, armnn::DataType::Float32>(DataLayout::NHWC);
-}
-
-BOOST_AUTO_TEST_CASE(CreateResizeBilinearFloat16NhwcWorkload)
-{
-    ClResizeBilinearWorkloadTest<ClResizeBilinearFloatWorkload, armnn::DataType::Float16>(DataLayout::NHWC);
-}
-
-BOOST_AUTO_TEST_SUITE_END()
diff --git a/src/backends/test/CreateWorkloadNeon.cpp b/src/backends/test/CreateWorkloadNeon.cpp
deleted file mode 100644
index b2ec563a69..0000000000
--- a/src/backends/test/CreateWorkloadNeon.cpp
+++ /dev/null
@@ -1,530 +0,0 @@
-//
-// Copyright © 2017 Arm Ltd. All rights reserved.
-// SPDX-License-Identifier: MIT
-//
-#include <backends/neon/NeonWorkloadFactory.hpp>
-#include <backends/neon/NeonTensorHandle.hpp>
-#include <backends/neon/workloads/NeonWorkloadUtils.hpp>
-#include <backends/neon/workloads/NeonWorkloads.hpp>
-#include <backends/MemCopyWorkload.hpp>
-
-#include "test/CreateWorkloadClNeon.hpp"
-
-BOOST_AUTO_TEST_SUITE(CreateWorkloadNeon)
-
-namespace
-{
-
-bool TestNeonTensorHandleInfo(armnn::INeonTensorHandle* handle, const armnn::TensorInfo& expectedInfo)
-{
-    using namespace armnn::armcomputetensorutils;
-
-    const arm_compute::ITensorInfo* handleInfo = handle->GetTensor().info();
-    const arm_compute::TensorInfo expectedAclInfo = BuildArmComputeTensorInfo(expectedInfo);
-
-    if (handleInfo->data_type() != expectedAclInfo.data_type())
-    {
-        return false;
-    }
-
-    if (handleInfo->num_dimensions() != expectedAclInfo.num_dimensions())
-    {
-        return false;
-    }
-
-    if (handleInfo->quantization_info() != expectedAclInfo.quantization_info())
-    {
-        return false;
-    }
-
-    for (std::size_t d = 0; d < expectedAclInfo.num_dimensions(); ++d)
-    {
-        if (handleInfo->dimension(d) != expectedAclInfo.dimension(d))
-        {
-            return false;
-        }
-    }
-
-    return true;
-}
-
-} // namespace
-
-template <typename ActivationWorkloadType, typename armnn::DataType DataType>
-static void NeonCreateActivationWorkloadTest()
-{
-    Graph graph;
-    NeonWorkloadFactory factory;
-    auto workload = CreateActivationWorkloadTest<ActivationWorkloadType, DataType>
-            (factory, graph);
-
-    // Checks that inputs/outputs are as we expect them (see definition of CreateActivationWorkloadTest).
-    ActivationQueueDescriptor queueDescriptor = workload->GetData();
-    auto inputHandle  = boost::polymorphic_downcast<INeonTensorHandle*>(queueDescriptor.m_Inputs[0]);
-    auto outputHandle = boost::polymorphic_downcast<INeonTensorHandle*>(queueDescriptor.m_Outputs[0]);
-    BOOST_TEST(TestNeonTensorHandleInfo(inputHandle, TensorInfo({1, 1}, DataType)));
-    BOOST_TEST(TestNeonTensorHandleInfo(outputHandle, TensorInfo({1, 1}, DataType)));
-}
-
-#ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
-BOOST_AUTO_TEST_CASE(CreateActivationFloat16Workload)
-{
-    NeonCreateActivationWorkloadTest<NeonActivationFloatWorkload, DataType::Float16>();
-}
-#endif
-
-BOOST_AUTO_TEST_CASE(CreateActivationFloatWorkload)
-{
-    NeonCreateActivationWorkloadTest<NeonActivationFloatWorkload, DataType::Float32>();
-}
-
-template <typename WorkloadType,
-          typename DescriptorType,
-          typename LayerType,
-          armnn::DataType DataType>
-static void NeonCreateArithmethicWorkloadTest()
-{
-    Graph graph;
-    NeonWorkloadFactory factory;
-    auto workload = CreateArithmeticWorkloadTest<WorkloadType, DescriptorType, LayerType, DataType>(factory, graph);
-
-    DescriptorType queueDescriptor = workload->GetData();
-    auto inputHandle1 = boost::polymorphic_downcast<INeonTensorHandle*>(queueDescriptor.m_Inputs[0]);
-    auto inputHandle2 = boost::polymorphic_downcast<INeonTensorHandle*>(queueDescriptor.m_Inputs[1]);
-    auto outputHandle = boost::polymorphic_downcast<INeonTensorHandle*>(queueDescriptor.m_Outputs[0]);
-    BOOST_TEST(TestNeonTensorHandleInfo(inputHandle1, TensorInfo({2, 3}, DataType)));
-    BOOST_TEST(TestNeonTensorHandleInfo(inputHandle2, TensorInfo({2, 3}, DataType)));
-    BOOST_TEST(TestNeonTensorHandleInfo(outputHandle, TensorInfo({2, 3}, DataType)));
-}
-
-#ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
-BOOST_AUTO_TEST_CASE(CreateAdditionFloat16Workload)
-{
-    NeonCreateArithmethicWorkloadTest<NeonAdditionFloatWorkload,
-                                      AdditionQueueDescriptor,
-                                      AdditionLayer,
-                                      DataType::Float16>();
-}
-#endif
-
-BOOST_AUTO_TEST_CASE(CreateAdditionFloatWorkload)
-{
-    NeonCreateArithmethicWorkloadTest<NeonAdditionFloatWorkload,
-                                      AdditionQueueDescriptor,
-                                      AdditionLayer,
-                                      DataType::Float32>();
-}
-
-#ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
-BOOST_AUTO_TEST_CASE(CreateSubtractionFloat16Workload)
-{
-    NeonCreateArithmethicWorkloadTest<NeonSubtractionFloatWorkload,
-                                      SubtractionQueueDescriptor,
-                                      SubtractionLayer,
-                                      DataType::Float16>();
-}
-#endif
-
-BOOST_AUTO_TEST_CASE(CreateSubtractionFloatWorkload)
-{
-    NeonCreateArithmethicWorkloadTest<NeonSubtractionFloatWorkload,
-                                      SubtractionQueueDescriptor,
-                                      SubtractionLayer,
-                                      DataType::Float32>();
-}
-
-#ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
-BOOST_AUTO_TEST_CASE(CreateMultiplicationFloat16Workload)
-{
-    NeonCreateArithmethicWorkloadTest<NeonMultiplicationFloatWorkload,
-                                      MultiplicationQueueDescriptor,
-                                      MultiplicationLayer,
-                                      DataType::Float16>();
-}
-#endif
-
-BOOST_AUTO_TEST_CASE(CreateMultiplicationFloatWorkload)
-{
-    NeonCreateArithmethicWorkloadTest<NeonMultiplicationFloatWorkload,
-                                      MultiplicationQueueDescriptor,
-                                      MultiplicationLayer,
-                                      DataType::Float32>();
-}
-
-template <typename BatchNormalizationWorkloadType, typename armnn::DataType DataType>
-static void NeonCreateBatchNormalizationWorkloadTest()
-{
-    Graph                graph;
-    NeonWorkloadFactory  factory;
-    auto workload = CreateBatchNormalizationWorkloadTest<BatchNormalizationWorkloadType, DataType>(factory, graph);
-
-    // Checks that outputs and inputs are as we expect them (see definition of CreateBatchNormalizationWorkloadTest).
-    BatchNormalizationQueueDescriptor queueDescriptor = workload->GetData();
-    auto inputHandle  = boost::polymorphic_downcast<INeonTensorHandle*>(queueDescriptor.m_Inputs[0]);
-    auto outputHandle = boost::polymorphic_downcast<INeonTensorHandle*>(queueDescriptor.m_Outputs[0]);
-    BOOST_TEST(TestNeonTensorHandleInfo(inputHandle, TensorInfo({2, 3, 1, 1}, DataType)));
-    BOOST_TEST(TestNeonTensorHandleInfo(outputHandle, TensorInfo({2, 3, 1, 1}, DataType)));
-}
-
-#ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
-BOOST_AUTO_TEST_CASE(CreateBatchNormalizationFloat16Workload)
-{
-    NeonCreateBatchNormalizationWorkloadTest<NeonBatchNormalizationFloatWorkload, DataType::Float16>();
-}
-#endif
-
-BOOST_AUTO_TEST_CASE(CreateBatchNormalizationFloatWorkload)
-{
-    NeonCreateBatchNormalizationWorkloadTest<NeonBatchNormalizationFloatWorkload, DataType::Float32>();
-}
-
-template <typename Convolution2dWorkloadType, typename armnn::DataType DataType>
-static void NeonCreateConvolution2dWorkloadTest(DataLayout dataLayout = DataLayout::NCHW)
-{
-    Graph                graph;
-    NeonWorkloadFactory  factory;
-    auto                 workload = CreateConvolution2dWorkloadTest<Convolution2dWorkloadType,
-                                    DataType>(factory, graph, dataLayout);
-
-    TensorShape inputShape  = (dataLayout == DataLayout::NCHW) ? TensorShape{2, 3, 8, 16} : TensorShape{2, 8, 16, 3};
-    TensorShape outputShape = (dataLayout == DataLayout::NCHW) ? TensorShape{2, 2, 2, 10} : TensorShape{2, 2, 10, 2};
-
-    // Checks that outputs and inputs are as we expect them (see definition of CreateConvolution2dWorkloadTest).
-    Convolution2dQueueDescriptor queueDescriptor = workload->GetData();
-    auto inputHandle  = boost::polymorphic_downcast<INeonTensorHandle*>(queueDescriptor.m_Inputs[0]);
-    auto outputHandle = boost::polymorphic_downcast<INeonTensorHandle*>(queueDescriptor.m_Outputs[0]);
-    BOOST_TEST(TestNeonTensorHandleInfo(inputHandle, TensorInfo(inputShape, DataType)));
-    BOOST_TEST(TestNeonTensorHandleInfo(outputHandle,  TensorInfo(outputShape, DataType)));
-}
-
-#ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
-BOOST_AUTO_TEST_CASE(CreateConvolution2dFloat16NchwWorkload)
-{
-    NeonCreateConvolution2dWorkloadTest<NeonConvolution2dFloatWorkload, DataType::Float16>();
-}
-
-BOOST_AUTO_TEST_CASE(CreateConvolution2dFloat16NhwcWorkload)
-{
-    NeonCreateConvolution2dWorkloadTest<NeonConvolution2dFloatWorkload, DataType::Float16>(DataLayout::NHWC);
-}
-
-#endif
-BOOST_AUTO_TEST_CASE(CreateConvolution2dFloatNchwWorkload)
-{
-    NeonCreateConvolution2dWorkloadTest<NeonConvolution2dFloatWorkload, DataType::Float32>();
-}
-
-BOOST_AUTO_TEST_CASE(CreateConvolution2dFloatNhwcWorkload)
-{
-    NeonCreateConvolution2dWorkloadTest<NeonConvolution2dFloatWorkload, DataType::Float32>(DataLayout::NHWC);
-}
-
-template <typename FullyConnectedWorkloadType, typename armnn::DataType DataType>
-static void NeonCreateFullyConnectedWorkloadTest()
-{
-    Graph               graph;
-    NeonWorkloadFactory factory;
-    auto                workload = CreateFullyConnectedWorkloadTest<FullyConnectedWorkloadType,
-                                   DataType>(factory, graph);
-
-    // Checks that outputs and inputs are as we expect them (see definition of CreateFullyConnectedWorkloadTest).
-    FullyConnectedQueueDescriptor queueDescriptor = workload->GetData();
-    auto inputHandle  = boost::polymorphic_downcast<INeonTensorHandle*>(queueDescriptor.m_Inputs[0]);
-    auto outputHandle = boost::polymorphic_downcast<INeonTensorHandle*>(queueDescriptor.m_Outputs[0]);
-    BOOST_TEST(TestNeonTensorHandleInfo(inputHandle, TensorInfo({3, 1, 4, 5}, DataType)));
-    BOOST_TEST(TestNeonTensorHandleInfo(outputHandle, TensorInfo({3, 7}, DataType)));
-}
-
-#ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
-BOOST_AUTO_TEST_CASE(CreateFullyConnectedFloat16Workload)
-{
-    NeonCreateFullyConnectedWorkloadTest<NeonFullyConnectedWorkload, DataType::Float16>();
-}
-#endif
-
-BOOST_AUTO_TEST_CASE(CreateFullyConnectedFloatWorkload)
-{
-    NeonCreateFullyConnectedWorkloadTest<NeonFullyConnectedWorkload, DataType::Float32>();
-}
-
-template <typename NormalizationWorkloadType, typename armnn::DataType DataType>
-static void NeonCreateNormalizationWorkloadTest(DataLayout dataLayout)
-{
-    Graph graph;
-    NeonWorkloadFactory factory;
-    auto workload = CreateNormalizationWorkloadTest<NormalizationWorkloadType, DataType>(factory, graph, dataLayout);
-
-    // Checks that outputs and inputs are as we expect them (see definition of CreateNormalizationWorkloadTest).
-    NormalizationQueueDescriptor queueDescriptor = workload->GetData();
-    auto inputHandle  = boost::polymorphic_downcast<INeonTensorHandle*>(queueDescriptor.m_Inputs[0]);
-    auto outputHandle = boost::polymorphic_downcast<INeonTensorHandle*>(queueDescriptor.m_Outputs[0]);
-    BOOST_TEST(TestNeonTensorHandleInfo(inputHandle, TensorInfo({3, 5, 5, 1}, DataType)));
-    BOOST_TEST(TestNeonTensorHandleInfo(outputHandle, TensorInfo({3, 5, 5, 1}, DataType)));
-}
-
-#ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
-BOOST_AUTO_TEST_CASE(CreateNormalizationFloat16NchwWorkload)
-{
-    NeonCreateNormalizationWorkloadTest<NeonNormalizationFloatWorkload, DataType::Float16>(DataLayout::NCHW);
-}
-
-BOOST_AUTO_TEST_CASE(CreateNormalizationFloat16NhwcWorkload)
-{
-    NeonCreateNormalizationWorkloadTest<NeonNormalizationFloatWorkload, DataType::Float16>(DataLayout::NHWC);
-}
-#endif
-
-BOOST_AUTO_TEST_CASE(CreateNormalizationFloatNchwWorkload)
-{
-    NeonCreateNormalizationWorkloadTest<NeonNormalizationFloatWorkload, DataType::Float32>(DataLayout::NCHW);
-}
-
-BOOST_AUTO_TEST_CASE(CreateNormalizationFloatNhwcWorkload)
-{
-    NeonCreateNormalizationWorkloadTest<NeonNormalizationFloatWorkload, DataType::Float32>(DataLayout::NHWC);
-}
-
-
-template <typename Pooling2dWorkloadType, typename armnn::DataType DataType>
-static void NeonCreatePooling2dWorkloadTest(DataLayout dataLayout = DataLayout::NCHW)
-{
-    Graph               graph;
-    NeonWorkloadFactory factory;
-    auto                workload = CreatePooling2dWorkloadTest<Pooling2dWorkloadType, DataType>
-                                   (factory, graph, dataLayout);
-
-    TensorShape inputShape  = (dataLayout == DataLayout::NCHW) ? TensorShape{3, 2, 5, 5} : TensorShape{3, 5, 5, 2};
-    TensorShape outputShape = (dataLayout == DataLayout::NCHW) ? TensorShape{3, 2, 2, 4} : TensorShape{3, 2, 4, 2};
-
-    // Checks that outputs and inputs are as we expect them (see definition of CreatePooling2dWorkloadTest).
-    Pooling2dQueueDescriptor queueDescriptor = workload->GetData();
-    auto inputHandle  = boost::polymorphic_downcast<INeonTensorHandle*>(queueDescriptor.m_Inputs[0]);
-    auto outputHandle = boost::polymorphic_downcast<INeonTensorHandle*>(queueDescriptor.m_Outputs[0]);
-    BOOST_TEST(TestNeonTensorHandleInfo(inputHandle, TensorInfo(inputShape, DataType)));
-    BOOST_TEST(TestNeonTensorHandleInfo(outputHandle, TensorInfo(outputShape, DataType)));
-}
-
-#ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
-BOOST_AUTO_TEST_CASE(CreatePooling2dFloat16Workload)
-{
-    NeonCreatePooling2dWorkloadTest<NeonPooling2dFloatWorkload, DataType::Float16>();
-}
-#endif
-
-BOOST_AUTO_TEST_CASE(CreatePooling2dFloatNchwWorkload)
-{
-    NeonCreatePooling2dWorkloadTest<NeonPooling2dFloatWorkload, DataType::Float32>(DataLayout::NCHW);
-}
-
-BOOST_AUTO_TEST_CASE(CreatePooling2dFloatNhwcWorkload)
-{
-    NeonCreatePooling2dWorkloadTest<NeonPooling2dFloatWorkload, DataType::Float32>(DataLayout::NHWC);
-}
-
-BOOST_AUTO_TEST_CASE(CreatePooling2dUint8NchwWorkload)
-{
-    NeonCreatePooling2dWorkloadTest<NeonPooling2dUint8Workload, DataType::QuantisedAsymm8>(DataLayout::NCHW);
-}
-
-BOOST_AUTO_TEST_CASE(CreatePooling2dUint8NhwcWorkload)
-{
-    NeonCreatePooling2dWorkloadTest<NeonPooling2dUint8Workload, DataType::QuantisedAsymm8>(DataLayout::NHWC);
-}
-
-template <typename ReshapeWorkloadType, typename armnn::DataType DataType>
-static void NeonCreateReshapeWorkloadTest()
-{
-    Graph               graph;
-    NeonWorkloadFactory factory;
-    auto                workload = CreateReshapeWorkloadTest<ReshapeWorkloadType, DataType>(factory, graph);
-
-    // Checks that outputs and inputs are as we expect them (see definition of CreateReshapeWorkloadTest).
-    ReshapeQueueDescriptor queueDescriptor = workload->GetData();
-    auto inputHandle  = boost::polymorphic_downcast<INeonTensorHandle*>(queueDescriptor.m_Inputs[0]);
-    auto outputHandle = boost::polymorphic_downcast<INeonTensorHandle*>(queueDescriptor.m_Outputs[0]);
-    BOOST_TEST(TestNeonTensorHandleInfo(inputHandle, TensorInfo({4, 1}, DataType)));
-    BOOST_TEST(TestNeonTensorHandleInfo(outputHandle, TensorInfo({1, 4}, DataType)));
-}
-
-#ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
-BOOST_AUTO_TEST_CASE(CreateReshapeFloat16Workload)
-{
-    NeonCreateReshapeWorkloadTest<NeonReshapeFloatWorkload, DataType::Float16>();
-}
-#endif
-
-BOOST_AUTO_TEST_CASE(CreateReshapeFloatWorkload)
-{
-    NeonCreateReshapeWorkloadTest<NeonReshapeFloatWorkload, DataType::Float32>();
-}
-
-BOOST_AUTO_TEST_CASE(CreateReshapeUint8Workload)
-{
-    NeonCreateReshapeWorkloadTest<NeonReshapeUint8Workload, DataType::QuantisedAsymm8>();
-}
-
-template <typename SoftmaxWorkloadType, typename armnn::DataType DataType>
-static void NeonCreateSoftmaxWorkloadTest()
-{
-    Graph               graph;
-    NeonWorkloadFactory factory;
-    auto workload = CreateSoftmaxWorkloadTest<SoftmaxWorkloadType, DataType>(factory, graph);
-
-    // Checks that outputs and inputs are as we expect them (see definition of CreateSoftmaxWorkloadTest).
-    SoftmaxQueueDescriptor queueDescriptor = workload->GetData();
-    auto inputHandle  = boost::polymorphic_downcast<INeonTensorHandle*>(queueDescriptor.m_Inputs[0]);
-    auto outputHandle = boost::polymorphic_downcast<INeonTensorHandle*>(queueDescriptor.m_Outputs[0]);
-    BOOST_TEST(TestNeonTensorHandleInfo(inputHandle, TensorInfo({4, 1}, DataType)));
-    BOOST_TEST(TestNeonTensorHandleInfo(outputHandle, TensorInfo({4, 1}, DataType)));
-}
-
-#ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
-BOOST_AUTO_TEST_CASE(CreateSoftmaxFloat16Workload)
-{
-    NeonCreateSoftmaxWorkloadTest<NeonSoftmaxFloatWorkload, DataType::Float16>();
-}
-#endif
-
-BOOST_AUTO_TEST_CASE(CreateSoftmaxFloatWorkload)
-{
-    NeonCreateSoftmaxWorkloadTest<NeonSoftmaxFloatWorkload, DataType::Float32>();
-}
-
-BOOST_AUTO_TEST_CASE(CreateSplitterWorkload)
-{
-    Graph graph;
-    NeonWorkloadFactory factory;
-    auto workload = CreateSplitterWorkloadTest<NeonSplitterFloatWorkload, DataType::Float32>(factory, graph);
-
-    // Checks that outputs are as we expect them (see definition of CreateSplitterWorkloadTest).
-    SplitterQueueDescriptor queueDescriptor = workload->GetData();
-    auto inputHandle = boost::polymorphic_downcast<INeonTensorHandle*>(queueDescriptor.m_Inputs[0]);
-    BOOST_TEST(TestNeonTensorHandleInfo(inputHandle, TensorInfo({5, 7, 7}, DataType::Float32)));
-
-    auto outputHandle0 = boost::polymorphic_downcast<INeonTensorHandle*>(queueDescriptor.m_Outputs[0]);
-    BOOST_TEST(TestNeonTensorHandleInfo(outputHandle0, TensorInfo({1, 7, 7}, DataType::Float32)));
-
-    auto outputHandle1 = boost::polymorphic_downcast<INeonTensorHandle*>(queueDescriptor.m_Outputs[1]);
-    BOOST_TEST(TestNeonTensorHandleInfo(outputHandle1, TensorInfo({2, 7, 7}, DataType::Float32)));
-
-    auto outputHandle2 = boost::polymorphic_downcast<INeonTensorHandle*>(queueDescriptor.m_Outputs[2]);
-    BOOST_TEST(TestNeonTensorHandleInfo(outputHandle2, TensorInfo({2, 7, 7}, DataType::Float32)));
-}
-
-BOOST_AUTO_TEST_CASE(CreateSplitterMerger)
-{
-    // Tests that it is possible to decide which output of the splitter layer
-    // should be lined to which input of the merger layer.
-    // We tested that is is possible to specify 0th output
-    // of the splitter to be the 1st input to the merger, and the 1st output of the splitter to be 0th input
-    // of the merger.
-
-    Graph graph;
-    NeonWorkloadFactory factory;
-
-    auto workloads =
-        CreateSplitterMergerWorkloadTest<NeonSplitterFloatWorkload, NeonMergerFloatWorkload,
-            DataType::Float32>(factory, graph);
-
-    auto wlSplitter = std::move(workloads.first);
-    auto wlMerger = std::move(workloads.second);
-
-    //Checks that the index of inputs/outputs matches what we declared on InputDescriptor construction.
-    armnn::INeonTensorHandle* sOut0 = dynamic_cast<armnn::INeonTensorHandle*>(wlSplitter->GetData().m_Outputs[0]);
-    armnn::INeonTensorHandle* sOut1 = dynamic_cast<armnn::INeonTensorHandle*>(wlSplitter->GetData().m_Outputs[1]);
-    armnn::INeonTensorHandle* mIn0 = dynamic_cast<armnn::INeonTensorHandle*>(wlMerger->GetData().m_Inputs[0]);
-    armnn::INeonTensorHandle* mIn1 = dynamic_cast<armnn::INeonTensorHandle*>(wlMerger->GetData().m_Inputs[1]);
-
-    BOOST_TEST(sOut0);
-    BOOST_TEST(sOut1);
-    BOOST_TEST(mIn0);
-    BOOST_TEST(mIn1);
-
-    bool validDataPointers = (sOut0 == mIn1) && (sOut1 == mIn0);
-
-    BOOST_TEST(validDataPointers);
-}
-
-BOOST_AUTO_TEST_CASE(CreateSingleOutputMultipleInputs)
-{
-    // Tests that it is possible to assign multiple (two) different layers to each of the outputs of a splitter layer.
-    // We created a splitter with two outputs. That each of those outputs is used by two different activation layers
-
-    Graph graph;
-    NeonWorkloadFactory factory;
-    std::unique_ptr<NeonSplitterFloatWorkload> wlSplitter;
-    std::unique_ptr<NeonActivationFloatWorkload> wlActiv0_0;
-    std::unique_ptr<NeonActivationFloatWorkload> wlActiv0_1;
-    std::unique_ptr<NeonActivationFloatWorkload> wlActiv1_0;
-    std::unique_ptr<NeonActivationFloatWorkload> wlActiv1_1;
-
-    CreateSplitterMultipleInputsOneOutputWorkloadTest<NeonSplitterFloatWorkload,
-        NeonActivationFloatWorkload, DataType::Float32>(factory, graph, wlSplitter, wlActiv0_0, wlActiv0_1,
-                                                                 wlActiv1_0, wlActiv1_1);
-
-    armnn::INeonTensorHandle* sOut0 = dynamic_cast<armnn::INeonTensorHandle*>(wlSplitter->GetData().m_Outputs[0]);
-    armnn::INeonTensorHandle* sOut1 = dynamic_cast<armnn::INeonTensorHandle*>(wlSplitter->GetData().m_Outputs[1]);
-    armnn::INeonTensorHandle* activ0_0Im = dynamic_cast<armnn::INeonTensorHandle*>(wlActiv0_0->GetData().m_Inputs[0]);
-    armnn::INeonTensorHandle* activ0_1Im = dynamic_cast<armnn::INeonTensorHandle*>(wlActiv0_1->GetData().m_Inputs[0]);
-    armnn::INeonTensorHandle* activ1_0Im = dynamic_cast<armnn::INeonTensorHandle*>(wlActiv1_0->GetData().m_Inputs[0]);
-    armnn::INeonTensorHandle* activ1_1Im = dynamic_cast<armnn::INeonTensorHandle*>(wlActiv1_1->GetData().m_Inputs[0]);
-
-
-    BOOST_TEST(sOut0);
-    BOOST_TEST(sOut1);
-    BOOST_TEST(activ0_0Im);
-    BOOST_TEST(activ0_1Im);
-    BOOST_TEST(activ1_0Im);
-    BOOST_TEST(activ1_1Im);
-
-    bool validDataPointers = (sOut0 == activ0_0Im) && (sOut0 == activ0_1Im) &&
-                             (sOut1 == activ1_0Im) && (sOut1 == activ1_1Im);
-
-    BOOST_TEST(validDataPointers);
-}
-
-BOOST_AUTO_TEST_CASE(CreateMemCopyWorkloadsNeon)
-{
-    NeonWorkloadFactory    factory;
-    CreateMemCopyWorkloads<INeonTensorHandle>(factory);
-}
-
-template <typename L2NormalizationWorkloadType, typename armnn::DataType DataType>
-static void NeonCreateL2NormalizationWorkloadTest(DataLayout dataLayout)
-{
-    Graph                graph;
-    NeonWorkloadFactory  factory;
-    auto                 workload = CreateL2NormalizationWorkloadTest<L2NormalizationWorkloadType,
-                                    DataType>(factory, graph, dataLayout);
-
-    // Checks that inputs/outputs are as we expect them (see definition of CreateNormalizationWorkloadTest).
-    L2NormalizationQueueDescriptor queueDescriptor = workload->GetData();
-    auto inputHandle = boost::polymorphic_downcast<INeonTensorHandle*>(queueDescriptor.m_Inputs[0]);
-    auto outputHandle = boost::polymorphic_downcast<INeonTensorHandle*>(queueDescriptor.m_Outputs[0]);
-    BOOST_TEST(TestNeonTensorHandleInfo(inputHandle, TensorInfo({ 5, 20, 50, 67 }, DataType)));
-    BOOST_TEST(TestNeonTensorHandleInfo(outputHandle, TensorInfo({ 5, 20, 50, 67 }, DataType)));
-}
-
-#ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
-BOOST_AUTO_TEST_CASE(CreateL2NormalizationFloat16NchwWorkload)
-{
-    NeonCreateL2NormalizationWorkloadTest<NeonL2NormalizationFloatWorkload, DataType::Float16>(DataLayout::NCHW);
-}
-
-BOOST_AUTO_TEST_CASE(CreateL2NormalizationFloat16NhwcWorkload)
-{
-    NeonCreateL2NormalizationWorkloadTest<NeonL2NormalizationFloatWorkload, DataType::Float16>(DataLayout::NHWC);
-}
-#endif
-
-BOOST_AUTO_TEST_CASE(CreateL2NormalizationNchwWorkload)
-{
-    NeonCreateL2NormalizationWorkloadTest<NeonL2NormalizationFloatWorkload, DataType::Float32>(DataLayout::NCHW);
-}
-
-BOOST_AUTO_TEST_CASE(CreateL2NormalizationNhwcWorkload)
-{
-    NeonCreateL2NormalizationWorkloadTest<NeonL2NormalizationFloatWorkload, DataType::Float32>(DataLayout::NHWC);
-}
-
-BOOST_AUTO_TEST_SUITE_END()
diff --git a/src/backends/test/CreateWorkloadRef.cpp b/src/backends/test/CreateWorkloadRef.cpp
deleted file mode 100644
index c30093da92..0000000000
--- a/src/backends/test/CreateWorkloadRef.cpp
+++ /dev/null
@@ -1,483 +0,0 @@
-//
-// Copyright © 2017 Arm Ltd. All rights reserved.
-// SPDX-License-Identifier: MIT
-//
-#include <backends/reference/RefWorkloadFactory.hpp>
-#include <backends/reference/workloads/RefWorkloads.hpp>
-#include <backends/CpuTensorHandle.hpp>
-
-#include "test/CreateWorkload.hpp"
-
-namespace
-{
-
-template<typename Workload>
-void CheckInputOutput(std::unique_ptr<Workload> workload, const TensorInfo& inputInfo, const TensorInfo& outputInfo)
-{
-    auto queueDescriptor = workload->GetData();
-    auto inputHandle  = boost::polymorphic_downcast<ConstCpuTensorHandle*>(queueDescriptor.m_Inputs[0]);
-    auto outputHandle = boost::polymorphic_downcast<CpuTensorHandle*>(queueDescriptor.m_Outputs[0]);
-    BOOST_TEST((inputHandle->GetTensorInfo() == inputInfo));
-    BOOST_TEST((outputHandle->GetTensorInfo() == outputInfo));
-}
-
-template <typename Workload>
-void CheckInputsOutput(std::unique_ptr<Workload> workload,
-                       const TensorInfo&         inputInfo0,
-                       const TensorInfo&         inputInfo1,
-                       const TensorInfo&         outputInfo)
-{
-    auto queueDescriptor = workload->GetData();
-    auto inputHandle0     = boost::polymorphic_downcast<ConstCpuTensorHandle*>(queueDescriptor.m_Inputs[0]);
-    auto inputHandle1     = boost::polymorphic_downcast<ConstCpuTensorHandle*>(queueDescriptor.m_Inputs[1]);
-    auto outputHandle    = boost::polymorphic_downcast<CpuTensorHandle*>(queueDescriptor.m_Outputs[0]);
-    BOOST_TEST((inputHandle0->GetTensorInfo() == inputInfo0));
-    BOOST_TEST((inputHandle1->GetTensorInfo() == inputInfo1));
-    BOOST_TEST((outputHandle->GetTensorInfo() == outputInfo));
-}
-}
-
-BOOST_AUTO_TEST_SUITE(CreateWorkloadRef)
-
-template <typename ActivationWorkloadType, armnn::DataType DataType>
-static void RefCreateActivationWorkloadTest()
-{
-    Graph graph;
-    RefWorkloadFactory factory;
-    auto workload = CreateActivationWorkloadTest<ActivationWorkloadType, DataType>(factory, graph);
-
-    // Checks that outputs are as we expect them (see definition of CreateActivationWorkloadTest).
-    CheckInputOutput(std::move(workload),
-        TensorInfo({ 1, 1 }, DataType),
-        TensorInfo({ 1, 1 }, DataType));
-}
-
-BOOST_AUTO_TEST_CASE(CreateActivationFloat32Workload)
-{
-    RefCreateActivationWorkloadTest<RefActivationFloat32Workload, armnn::DataType::Float32>();
-}
-
-BOOST_AUTO_TEST_CASE(CreateActivationUint8Workload)
-{
-    RefCreateActivationWorkloadTest<RefActivationUint8Workload, armnn::DataType::QuantisedAsymm8>();
-}
-
-template <typename WorkloadType,
-          typename DescriptorType,
-          typename LayerType,
-          armnn::DataType DataType>
-static void RefCreateArithmethicWorkloadTest()
-{
-    Graph graph;
-    RefWorkloadFactory factory;
-    auto workload = CreateArithmeticWorkloadTest<WorkloadType, DescriptorType, LayerType, DataType>(factory, graph);
-
-    CheckInputsOutput(std::move(workload),
-        TensorInfo({ 2, 3 }, DataType),
-        TensorInfo({ 2, 3 }, DataType),
-        TensorInfo({ 2, 3 }, DataType));
-}
-
-BOOST_AUTO_TEST_CASE(CreateAdditionFloatWorkload)
-{
-    RefCreateArithmethicWorkloadTest<RefAdditionFloat32Workload,
-                                     AdditionQueueDescriptor,
-                                     AdditionLayer,
-                                     armnn::DataType::Float32>();
-}
-
-BOOST_AUTO_TEST_CASE(CreateAdditionUint8Workload)
-{
-    RefCreateArithmethicWorkloadTest<RefAdditionUint8Workload,
-                                     AdditionQueueDescriptor,
-                                     AdditionLayer,
-                                     armnn::DataType::QuantisedAsymm8>();
-}
-
-BOOST_AUTO_TEST_CASE(CreateSubtractionFloatWorkload)
-{
-    RefCreateArithmethicWorkloadTest<RefSubtractionFloat32Workload,
-                                     SubtractionQueueDescriptor,
-                                     SubtractionLayer,
-                                     armnn::DataType::Float32>();
-}
-
-BOOST_AUTO_TEST_CASE(CreateSubtractionUint8Workload)
-{
-    RefCreateArithmethicWorkloadTest<RefSubtractionUint8Workload,
-                                     SubtractionQueueDescriptor,
-                                     SubtractionLayer,
-                                     armnn::DataType::QuantisedAsymm8>();
-}
-
-BOOST_AUTO_TEST_CASE(CreateMultiplicationFloatWorkload)
-{
-    RefCreateArithmethicWorkloadTest<RefMultiplicationFloat32Workload,
-                                     MultiplicationQueueDescriptor,
-                                     MultiplicationLayer,
-                                     armnn::DataType::Float32>();
-}
-
-BOOST_AUTO_TEST_CASE(CreateMultiplicationUint8Workload)
-{
-    RefCreateArithmethicWorkloadTest<RefMultiplicationUint8Workload,
-                                     MultiplicationQueueDescriptor,
-                                     MultiplicationLayer,
-                                     armnn::DataType::QuantisedAsymm8>();
-}
-
-BOOST_AUTO_TEST_CASE(CreateDivisionFloatWorkload)
-{
-    RefCreateArithmethicWorkloadTest<RefDivisionFloat32Workload,
-                                     DivisionQueueDescriptor,
-                                     DivisionLayer,
-                                     armnn::DataType::Float32>();
-}
-
-BOOST_AUTO_TEST_CASE(CreateDivisionUint8Workload)
-{
-    RefCreateArithmethicWorkloadTest<RefDivisionUint8Workload,
-                                     DivisionQueueDescriptor,
-                                     DivisionLayer,
-                                     armnn::DataType::QuantisedAsymm8>();
-}
-
-BOOST_AUTO_TEST_CASE(CreateBatchNormalizationWorkload)
-{
-    Graph                graph;
-    RefWorkloadFactory factory;
-    auto workload = CreateBatchNormalizationWorkloadTest<RefBatchNormalizationFloat32Workload, armnn::DataType::Float32>
-                    (factory, graph);
-
-    // Checks that outputs and inputs are as we expect them (see definition of CreateBatchNormalizationWorkloadTest).
-    CheckInputOutput(
-        std::move(workload), TensorInfo({2, 3, 1, 1}, DataType::Float32), TensorInfo({2, 3, 1, 1}, DataType::Float32));
-}
-
-BOOST_AUTO_TEST_CASE(CreateConvertFp16ToFp32Float32Workload)
-{
-    Graph                graph;
-    RefWorkloadFactory factory;
-    auto workload = CreateConvertFp16ToFp32WorkloadTest<RefConvertFp16ToFp32Workload>(factory, graph);
-
-    // Checks that outputs and inputs are as we expect them
-    CheckInputOutput(
-        std::move(workload), TensorInfo({1, 3, 2, 3}, DataType::Float16), TensorInfo({1, 3, 2, 3}, DataType::Float32));
-}
-
-BOOST_AUTO_TEST_CASE(CreateConvertFp32ToFp16Float16Workload)
-{
-    Graph                graph;
-    RefWorkloadFactory factory;
-    auto workload = CreateConvertFp32ToFp16WorkloadTest<RefConvertFp32ToFp16Workload>(factory, graph);
-
-    // Checks that outputs and inputs are as we expect them
-    CheckInputOutput(
-        std::move(workload), TensorInfo({1, 3, 2, 3}, DataType::Float32), TensorInfo({1, 3, 2, 3}, DataType::Float16));
-}
-
-BOOST_AUTO_TEST_CASE(CreateConvolution2dWorkload)
-{
-    Graph                graph;
-    RefWorkloadFactory factory;
-    auto                 workload = CreateConvolution2dWorkloadTest<RefConvolution2dFloat32Workload,
-                         DataType::Float32>(factory, graph);
-
-    // Checks that outputs and inputs are as we expect them (see definition of CreateConvolution2dWorkloadTest).
-    CheckInputOutput(std::move(workload),
-                     TensorInfo({2, 3, 8, 16}, DataType::Float32),
-                     TensorInfo({2, 2, 2, 10}, DataType::Float32));
-}
-
-BOOST_AUTO_TEST_CASE(CreateDepthwiseConvolution2dWorkload)
-{
-    Graph                graph;
-    RefWorkloadFactory factory;
-    auto                 workload =
-        CreateDepthwiseConvolution2dWorkloadTest<RefDepthwiseConvolution2dFloat32Workload>(factory, graph);
-
-    // Checks that outputs and inputs are as we expect them (see definition of CreateConvolution2dWorkloadTest).
-    CheckInputOutput(std::move(workload),
-                     TensorInfo({2, 3, 8, 16}, DataType::Float32),
-                     TensorInfo({2, 9, 2, 10}, DataType::Float32));
-}
-
-template <typename FullyConnectedWorkloadType, armnn::DataType DataType>
-static void RefCreateFullyConnectedWorkloadTest()
-{
-    Graph graph;
-    RefWorkloadFactory factory;
-    auto workload = CreateFullyConnectedWorkloadTest<FullyConnectedWorkloadType, DataType>(factory, graph);
-
-    // Checks that outputs and inputs are as we expect them (see definition of CreateFullyConnectedWorkloadTest).
-    float inputsQScale = DataType == armnn::DataType::QuantisedAsymm8 ? 1.0f : 0.0;
-    float outputQScale = DataType == armnn::DataType::QuantisedAsymm8 ? 2.0f : 0.0;
-    CheckInputOutput(std::move(workload),
-        TensorInfo({ 3, 1, 4, 5 }, DataType, inputsQScale),
-        TensorInfo({ 3, 7 }, DataType, outputQScale));
-}
-
-BOOST_AUTO_TEST_CASE(CreateFullyConnectedFloat32Workload)
-{
-    RefCreateFullyConnectedWorkloadTest<RefFullyConnectedFloat32Workload, armnn::DataType::Float32>();
-}
-
-BOOST_AUTO_TEST_CASE(CreateFullyConnectedUint8Workload)
-{
-    RefCreateFullyConnectedWorkloadTest<RefFullyConnectedUint8Workload, armnn::DataType::QuantisedAsymm8>();
-}
-
-template <typename NormalizationWorkloadType, armnn::DataType DataType>
-static void RefCreateNormalizationWorkloadTest()
-{
-    Graph graph;
-    RefWorkloadFactory factory;
-    auto workload = CreateNormalizationWorkloadTest<NormalizationWorkloadType, DataType>(factory, graph);
-
-    // Checks that outputs and inputs are as we expect them (see definition of CreateNormalizationWorkloadTest).
-    CheckInputOutput(std::move(workload),
-                     TensorInfo({3, 5, 5, 1}, DataType),
-                     TensorInfo({3, 5, 5, 1}, DataType));
-}
-
-BOOST_AUTO_TEST_CASE(CreateRefNormalizationNchwWorkload)
-{
-    RefCreateNormalizationWorkloadTest<RefNormalizationFloat32Workload, armnn::DataType::Float32>();
-}
-
-template <typename Pooling2dWorkloadType, armnn::DataType DataType>
-static void RefCreatePooling2dWorkloadTest()
-{
-    Graph graph;
-    RefWorkloadFactory factory;
-    auto workload = CreatePooling2dWorkloadTest<Pooling2dWorkloadType, DataType>(factory, graph);
-
-    // Checks that outputs and inputs are as we expect them (see definition of CreatePooling2dWorkloadTest).
-    CheckInputOutput(
-        std::move(workload),
-        TensorInfo({3, 2, 5, 5}, DataType),
-        TensorInfo({3, 2, 2, 4}, DataType));
-}
-
-BOOST_AUTO_TEST_CASE(CreatePooling2dFloat32Workload)
-{
-    RefCreatePooling2dWorkloadTest<RefPooling2dFloat32Workload, armnn::DataType::Float32>();
-}
-
-BOOST_AUTO_TEST_CASE(CreatePooling2dUint8Workload)
-{
-    RefCreatePooling2dWorkloadTest<RefPooling2dUint8Workload, armnn::DataType::QuantisedAsymm8>();
-}
-
-template <typename SoftmaxWorkloadType, armnn::DataType DataType>
-static void RefCreateSoftmaxWorkloadTest()
-{
-    Graph graph;
-    RefWorkloadFactory factory;
-    auto workload = CreateSoftmaxWorkloadTest<SoftmaxWorkloadType, DataType>(factory, graph);
-
-    // Checks that outputs and inputs are as we expect them (see definition of CreateSoftmaxWorkloadTest).
-    CheckInputOutput(
-        std::move(workload),
-        TensorInfo({4, 1}, DataType),
-        TensorInfo({4, 1}, DataType));
-}
-
-BOOST_AUTO_TEST_CASE(CreateSoftmaxFloat32Workload)
-{
-    RefCreateSoftmaxWorkloadTest<RefSoftmaxFloat32Workload, armnn::DataType::Float32>();
-}
-
-BOOST_AUTO_TEST_CASE(CreateSoftmaxUint8Workload)
-{
-    RefCreateSoftmaxWorkloadTest<RefSoftmaxUint8Workload, armnn::DataType::QuantisedAsymm8>();
-}
-
-template <typename SplitterWorkloadType, armnn::DataType DataType>
-static void RefCreateSplitterWorkloadTest()
-{
-    Graph graph;
-    RefWorkloadFactory factory;
-    auto workload = CreateSplitterWorkloadTest<SplitterWorkloadType, DataType>(factory, graph);
-
-    // Checks that outputs are as we expect them (see definition of CreateSplitterWorkloadTest).
-    SplitterQueueDescriptor queueDescriptor = workload->GetData();
-    auto inputHandle = boost::polymorphic_downcast<ConstCpuTensorHandle*>(queueDescriptor.m_Inputs[0]);
-    BOOST_TEST((inputHandle->GetTensorInfo() == TensorInfo({ 5, 7, 7 }, DataType)));
-
-    auto outputHandle0 = boost::polymorphic_downcast<CpuTensorHandle*>(queueDescriptor.m_Outputs[0]);
-    BOOST_TEST((outputHandle0->GetTensorInfo() == TensorInfo({ 1, 7, 7 }, DataType)));
-
-    auto outputHandle1 = boost::polymorphic_downcast<CpuTensorHandle*>(queueDescriptor.m_Outputs[1]);
-    BOOST_TEST((outputHandle1->GetTensorInfo() == TensorInfo({ 2, 7, 7 }, DataType)));
-
-    auto outputHandle2 = boost::polymorphic_downcast<CpuTensorHandle*>(queueDescriptor.m_Outputs[2]);
-    BOOST_TEST((outputHandle2->GetTensorInfo() == TensorInfo({ 2, 7, 7 }, DataType)));
-}
-
-BOOST_AUTO_TEST_CASE(CreateSplitterFloat32Workload)
-{
-    RefCreateSplitterWorkloadTest<RefSplitterFloat32Workload, armnn::DataType::Float32>();
-}
-
-BOOST_AUTO_TEST_CASE(CreateSplitterUint8Workload)
-{
-    RefCreateSplitterWorkloadTest<RefSplitterUint8Workload, armnn::DataType::QuantisedAsymm8>();
-}
-
-template <typename SplitterWorkloadType, typename MergerWorkloadType, armnn::DataType DataType>
-static void RefCreateSplitterMergerWorkloadTest()
-{
-    // Tests that it is possible to decide which output of the splitter layer
-    // should be lined to which input of the merger layer.
-    // We tested that is is possible to specify 0th output
-    // of the splitter to be the 1st input to the merger and the 1st output of the splitter to be 0th input
-    // of the merger.
-
-    Graph graph;
-    RefWorkloadFactory factory;
-    auto workloads = CreateSplitterMergerWorkloadTest<SplitterWorkloadType, MergerWorkloadType, DataType>
-        (factory, graph);
-
-    auto wlSplitter = std::move(workloads.first);
-    auto wlMerger = std::move(workloads.second);
-
-    //Checks that the index of inputs/outputs matches what we declared on InputDescriptor construction.
-    armnn::CpuTensorHandle* sOut0 = dynamic_cast<armnn::CpuTensorHandle*>(wlSplitter->GetData().m_Outputs[0]);
-    armnn::CpuTensorHandle* sOut1 = dynamic_cast<armnn::CpuTensorHandle*>(wlSplitter->GetData().m_Outputs[1]);
-    armnn::CpuTensorHandle* mIn0 = dynamic_cast<armnn::CpuTensorHandle*>(wlMerger->GetData().m_Inputs[0]);
-    armnn::CpuTensorHandle* mIn1 = dynamic_cast<armnn::CpuTensorHandle*>(wlMerger->GetData().m_Inputs[1]);
-
-    BOOST_TEST(sOut0);
-    BOOST_TEST(sOut1);
-    BOOST_TEST(mIn0);
-    BOOST_TEST(mIn1);
-
-    bool validDataPointers = (sOut0 == mIn1) && (sOut1 == mIn0);
-
-    BOOST_TEST(validDataPointers);
-}
-
-BOOST_AUTO_TEST_CASE(CreateSplitterMergerFloat32)
-{
-    RefCreateSplitterMergerWorkloadTest<RefSplitterFloat32Workload, RefMergerFloat32Workload, DataType::Float32>();
-}
-
-BOOST_AUTO_TEST_CASE(CreateSplitterMergerUint8)
-{
-    RefCreateSplitterMergerWorkloadTest<RefSplitterUint8Workload, RefMergerUint8Workload, DataType::QuantisedAsymm8>();
-}
-
-template <typename SplitterWorkloadType, typename ActivationWorkloadType, armnn::DataType DataType>
-static void RefCreateSingleOutputMultipleInputsTest()
-{
-    // Tests that it is possible to assign multiple (two) different layers to each of the outputs of a splitter layer.
-    // We created a splitter with two outputs. That each of those outputs is used by two different activation layers.
-
-    Graph graph;
-    RefWorkloadFactory factory;
-    std::unique_ptr<SplitterWorkloadType> wlSplitter;
-    std::unique_ptr<ActivationWorkloadType> wlActiv0_0;
-    std::unique_ptr<ActivationWorkloadType> wlActiv0_1;
-    std::unique_ptr<ActivationWorkloadType> wlActiv1_0;
-    std::unique_ptr<ActivationWorkloadType> wlActiv1_1;
-
-    CreateSplitterMultipleInputsOneOutputWorkloadTest<SplitterWorkloadType,
-        ActivationWorkloadType, DataType>(factory, graph, wlSplitter, wlActiv0_0, wlActiv0_1, wlActiv1_0, wlActiv1_1);
-
-    armnn::CpuTensorHandle* sOut0 = dynamic_cast<armnn::CpuTensorHandle*>(wlSplitter->GetData().m_Outputs[0]);
-    armnn::CpuTensorHandle* sOut1 = dynamic_cast<armnn::CpuTensorHandle*>(wlSplitter->GetData().m_Outputs[1]);
-    armnn::CpuTensorHandle* activ0_0Im = dynamic_cast<armnn::CpuTensorHandle*>(wlActiv0_0->GetData().m_Inputs[0]);
-    armnn::CpuTensorHandle* activ0_1Im = dynamic_cast<armnn::CpuTensorHandle*>(wlActiv0_1->GetData().m_Inputs[0]);
-    armnn::CpuTensorHandle* activ1_0Im = dynamic_cast<armnn::CpuTensorHandle*>(wlActiv1_0->GetData().m_Inputs[0]);
-    armnn::CpuTensorHandle* activ1_1Im = dynamic_cast<armnn::CpuTensorHandle*>(wlActiv1_1->GetData().m_Inputs[0]);
-
-
-    BOOST_TEST(sOut0);
-    BOOST_TEST(sOut1);
-    BOOST_TEST(activ0_0Im);
-    BOOST_TEST(activ0_1Im);
-    BOOST_TEST(activ1_0Im);
-    BOOST_TEST(activ1_1Im);
-
-    bool validDataPointers = (sOut0 == activ0_0Im) && (sOut0 == activ0_1Im) &&
-                             (sOut1 == activ1_0Im) && (sOut1 == activ1_1Im);
-
-    BOOST_TEST(validDataPointers);
-}
-
-BOOST_AUTO_TEST_CASE(CreateSingleOutputMultipleInputsFloat32)
-{
-    RefCreateSingleOutputMultipleInputsTest<RefSplitterFloat32Workload, RefActivationFloat32Workload,
-        armnn::DataType::Float32>();
-}
-
-BOOST_AUTO_TEST_CASE(CreateSingleOutputMultipleInputsUint8)
-{
-    RefCreateSingleOutputMultipleInputsTest<RefSplitterUint8Workload, RefActivationUint8Workload,
-        armnn::DataType::QuantisedAsymm8>();
-}
-
-template <typename ResizeBilinearWorkloadType, armnn::DataType DataType>
-static void RefCreateResizeBilinearTest()
-{
-    Graph graph;
-    RefWorkloadFactory factory;
-    auto workload = CreateResizeBilinearWorkloadTest<ResizeBilinearWorkloadType, DataType>(factory, graph);
-
-    // Checks that outputs and inputs are as we expect them (see definition of CreateResizeBilinearWorkloadTest).
-    CheckInputOutput(
-        std::move(workload),
-        TensorInfo({ 2, 3, 4, 4 }, DataType),
-        TensorInfo({ 2, 3, 2, 2 }, DataType));
-}
-
-BOOST_AUTO_TEST_CASE(CreateResizeBilinearFloat32)
-{
-    RefCreateResizeBilinearTest<RefResizeBilinearFloat32Workload, armnn::DataType::Float32>();
-}
-
-BOOST_AUTO_TEST_CASE(CreateResizeBilinearUint8)
-{
-    RefCreateResizeBilinearTest<RefResizeBilinearUint8Workload, armnn::DataType::QuantisedAsymm8>();
-}
-
-BOOST_AUTO_TEST_CASE(CreateL2NormalizationFloat32)
-{
-    Graph graph;
-    RefWorkloadFactory factory;
-    auto workload = CreateL2NormalizationWorkloadTest<RefL2NormalizationFloat32Workload, armnn::DataType::Float32>
-            (factory, graph);
-
-    // Checks that outputs and inputs are as we expect them (see definition of CreateL2NormalizationWorkloadTest).
-    CheckInputOutput(
-        std::move(workload),
-        TensorInfo({ 5, 20, 50, 67 }, armnn::DataType::Float32),
-        TensorInfo({ 5, 20, 50, 67 }, armnn::DataType::Float32));
-}
-
-template <typename ReshapeWorkloadType, armnn::DataType DataType>
-static void RefCreateReshapeWorkloadTest()
-{
-    Graph graph;
-    RefWorkloadFactory factory;
-    auto workload = CreateReshapeWorkloadTest<ReshapeWorkloadType, DataType>(factory, graph);
-
-    // Checks that outputs and inputs are as we expect them (see definition of CreateReshapeWorkloadTest).
-    CheckInputOutput(
-        std::move(workload),
-        TensorInfo({ 4, 1 }, DataType),
-        TensorInfo({ 1, 4 }, DataType));
-}
-
-BOOST_AUTO_TEST_CASE(CreateReshapeFloat32Workload)
-{
-    RefCreateReshapeWorkloadTest<RefReshapeFloat32Workload, armnn::DataType::Float32>();
-}
-
-BOOST_AUTO_TEST_CASE(CreateReshapeUint8Workload)
-{
-    RefCreateReshapeWorkloadTest<RefReshapeUint8Workload, armnn::DataType::QuantisedAsymm8>();
-}
-
-BOOST_AUTO_TEST_SUITE_END()
diff --git a/src/backends/test/IsLayerSupportedTest.cpp b/src/backends/test/IsLayerSupportedTest.cpp
deleted file mode 100644
index 089822dade..0000000000
--- a/src/backends/test/IsLayerSupportedTest.cpp
+++ /dev/null
@@ -1,239 +0,0 @@
-//
-// Copyright © 2017 Arm Ltd. All rights reserved.
-// SPDX-License-Identifier: MIT
-//
-#include <boost/test/unit_test.hpp>
-
-#include <test/TensorHelpers.hpp>
-#include "LayerTests.hpp"
-
-#include <backends/CpuTensorHandle.hpp>
-#include <backends/reference/RefWorkloadFactory.hpp>
-#include <backends/neon/NeonWorkloadFactory.hpp>
-
-#include <string>
-#include <iostream>
-#include <backends/cl/ClWorkloadFactory.hpp>
-
-#include "IsLayerSupportedTestImpl.hpp"
-#include <backends/test/ClContextControlFixture.hpp>
-
-#include <layers/ConvertFp16ToFp32Layer.hpp>
-#include <layers/ConvertFp32ToFp16Layer.hpp>
-
-BOOST_AUTO_TEST_SUITE(IsLayerSupported)
-
-BOOST_AUTO_TEST_CASE(IsLayerSupportedLayerTypeMatches)
-{
-    LayerTypeMatchesTest();
-}
-
-BOOST_AUTO_TEST_CASE(IsLayerSupportedFloat16Reference)
-{
-    armnn::RefWorkloadFactory factory;
-    IsLayerSupportedTests<armnn::RefWorkloadFactory, armnn::DataType::Float16>(&factory);
-}
-
-BOOST_AUTO_TEST_CASE(IsLayerSupportedFloat32Reference)
-{
-    armnn::RefWorkloadFactory factory;
-    IsLayerSupportedTests<armnn::RefWorkloadFactory, armnn::DataType::Float32>(&factory);
-}
-
-BOOST_AUTO_TEST_CASE(IsLayerSupportedUint8Reference)
-{
-    armnn::RefWorkloadFactory factory;
-    IsLayerSupportedTests<armnn::RefWorkloadFactory, armnn::DataType::QuantisedAsymm8>(&factory);
-}
-
-BOOST_AUTO_TEST_CASE(IsConvertFp16ToFp32SupportedReference)
-{
-    std::string reasonIfUnsupported;
-
-    bool result = IsConvertLayerSupportedTests<armnn::RefWorkloadFactory, armnn::ConvertFp16ToFp32Layer,
-      armnn::DataType::Float16, armnn::DataType::Float32>(reasonIfUnsupported);
-
-    BOOST_CHECK(result);
-}
-
-BOOST_AUTO_TEST_CASE(IsConvertFp16ToFp32SupportedFp32InputReference)
-{
-    std::string reasonIfUnsupported;
-
-    bool result = IsConvertLayerSupportedTests<armnn::RefWorkloadFactory, armnn::ConvertFp16ToFp32Layer,
-      armnn::DataType::Float32, armnn::DataType::Float32>(reasonIfUnsupported);
-
-    BOOST_CHECK(!result);
-    BOOST_CHECK_EQUAL(reasonIfUnsupported, "Layer is not supported with float32 data type input");
-}
-
-BOOST_AUTO_TEST_CASE(IsConvertFp16ToFp32SupportedFp16OutputReference)
-{
-    std::string reasonIfUnsupported;
-
-    bool result = IsConvertLayerSupportedTests<armnn::RefWorkloadFactory, armnn::ConvertFp16ToFp32Layer,
-      armnn::DataType::Float16, armnn::DataType::Float16>(reasonIfUnsupported);
-
-    BOOST_CHECK(!result);
-    BOOST_CHECK_EQUAL(reasonIfUnsupported, "Layer is not supported with float16 data type output");
-}
-
-BOOST_AUTO_TEST_CASE(IsConvertFp32ToFp16SupportedReference)
-{
-    std::string reasonIfUnsupported;
-
-    bool result = IsConvertLayerSupportedTests<armnn::RefWorkloadFactory, armnn::ConvertFp32ToFp16Layer,
-      armnn::DataType::Float32, armnn::DataType::Float16>(reasonIfUnsupported);
-
-    BOOST_CHECK(result);
-}
-
-BOOST_AUTO_TEST_CASE(IsConvertFp32ToFp16SupportedFp16InputReference)
-{
-    std::string reasonIfUnsupported;
-
-    bool result = IsConvertLayerSupportedTests<armnn::RefWorkloadFactory, armnn::ConvertFp32ToFp16Layer,
-      armnn::DataType::Float16, armnn::DataType::Float16>(reasonIfUnsupported);
-
-    BOOST_CHECK(!result);
-    BOOST_CHECK_EQUAL(reasonIfUnsupported, "Layer is not supported with float16 data type input");
-}
-
-BOOST_AUTO_TEST_CASE(IsConvertFp32ToFp16SupportedFp32OutputReference)
-{
-    std::string reasonIfUnsupported;
-
-    bool result = IsConvertLayerSupportedTests<armnn::RefWorkloadFactory, armnn::ConvertFp32ToFp16Layer,
-      armnn::DataType::Float32, armnn::DataType::Float32>(reasonIfUnsupported);
-
-    BOOST_CHECK(!result);
-    BOOST_CHECK_EQUAL(reasonIfUnsupported, "Layer is not supported with float32 data type output");
-}
-
-#ifdef ARMCOMPUTENEON_ENABLED
-BOOST_AUTO_TEST_CASE(IsLayerSupportedFloat16Neon)
-{
-    armnn::NeonWorkloadFactory factory;
-    IsLayerSupportedTests<armnn::NeonWorkloadFactory, armnn::DataType::Float16>(&factory);
-}
-
-BOOST_AUTO_TEST_CASE(IsLayerSupportedFloat32Neon)
-{
-    armnn::NeonWorkloadFactory factory;
-    IsLayerSupportedTests<armnn::NeonWorkloadFactory, armnn::DataType::Float32>(&factory);
-}
-
-BOOST_AUTO_TEST_CASE(IsLayerSupportedUint8Neon)
-{
-    armnn::NeonWorkloadFactory factory;
-    IsLayerSupportedTests<armnn::NeonWorkloadFactory, armnn::DataType::QuantisedAsymm8>(&factory);
-}
-
-BOOST_AUTO_TEST_CASE(IsConvertFp16ToFp32SupportedNeon)
-{
-    std::string reasonIfUnsupported;
-
-    bool result = IsConvertLayerSupportedTests<armnn::NeonWorkloadFactory, armnn::ConvertFp16ToFp32Layer,
-      armnn::DataType::Float16, armnn::DataType::Float32>(reasonIfUnsupported);
-
-    BOOST_CHECK(result);
-}
-
-BOOST_AUTO_TEST_CASE(IsConvertFp32ToFp16SupportedNeon)
-{
-    std::string reasonIfUnsupported;
-
-    bool result = IsConvertLayerSupportedTests<armnn::NeonWorkloadFactory, armnn::ConvertFp32ToFp16Layer,
-      armnn::DataType::Float32, armnn::DataType::Float16>(reasonIfUnsupported);
-
-    BOOST_CHECK(result);
-}
-#endif //#ifdef ARMCOMPUTENEON_ENABLED.
-
-
-#ifdef ARMCOMPUTECL_ENABLED
-
-BOOST_FIXTURE_TEST_CASE(IsLayerSupportedFloat16Cl, ClContextControlFixture)
-{
-    armnn::ClWorkloadFactory factory;
-    IsLayerSupportedTests<armnn::ClWorkloadFactory, armnn::DataType::Float16>(&factory);
-}
-
-BOOST_FIXTURE_TEST_CASE(IsLayerSupportedFloat32Cl, ClContextControlFixture)
-{
-    armnn::ClWorkloadFactory factory;
-    IsLayerSupportedTests<armnn::ClWorkloadFactory, armnn::DataType::Float32>(&factory);
-}
-
-BOOST_FIXTURE_TEST_CASE(IsLayerSupportedUint8Cl, ClContextControlFixture)
-{
-    armnn::ClWorkloadFactory factory;
-    IsLayerSupportedTests<armnn::ClWorkloadFactory, armnn::DataType::QuantisedAsymm8>(&factory);
-}
-
-BOOST_FIXTURE_TEST_CASE(IsConvertFp16ToFp32SupportedCl, ClContextControlFixture)
-{
-    std::string reasonIfUnsupported;
-
-    bool result = IsConvertLayerSupportedTests<armnn::ClWorkloadFactory, armnn::ConvertFp16ToFp32Layer,
-      armnn::DataType::Float16, armnn::DataType::Float32>(reasonIfUnsupported);
-
-    BOOST_CHECK(result);
-}
-
-BOOST_FIXTURE_TEST_CASE(IsConvertFp16ToFp32SupportedFp32InputCl, ClContextControlFixture)
-{
-    std::string reasonIfUnsupported;
-
-    bool result = IsConvertLayerSupportedTests<armnn::ClWorkloadFactory, armnn::ConvertFp16ToFp32Layer,
-      armnn::DataType::Float32, armnn::DataType::Float32>(reasonIfUnsupported);
-
-    BOOST_CHECK(!result);
-    BOOST_CHECK_EQUAL(reasonIfUnsupported, "Input should be Float16");
-}
-
-BOOST_FIXTURE_TEST_CASE(IsConvertFp16ToFp32SupportedFp16OutputCl, ClContextControlFixture)
-{
-    std::string reasonIfUnsupported;
-
-    bool result = IsConvertLayerSupportedTests<armnn::ClWorkloadFactory, armnn::ConvertFp16ToFp32Layer,
-      armnn::DataType::Float16, armnn::DataType::Float16>(reasonIfUnsupported);
-
-    BOOST_CHECK(!result);
-    BOOST_CHECK_EQUAL(reasonIfUnsupported, "Output should be Float32");
-}
-
-BOOST_FIXTURE_TEST_CASE(IsConvertFp32ToFp16SupportedCl, ClContextControlFixture)
-{
-    std::string reasonIfUnsupported;
-
-    bool result = IsConvertLayerSupportedTests<armnn::ClWorkloadFactory, armnn::ConvertFp32ToFp16Layer,
-      armnn::DataType::Float32, armnn::DataType::Float16>(reasonIfUnsupported);
-
-    BOOST_CHECK(result);
-}
-
-BOOST_FIXTURE_TEST_CASE(IsConvertFp32ToFp16SupportedFp16InputCl, ClContextControlFixture)
-{
-    std::string reasonIfUnsupported;
-
-    bool result = IsConvertLayerSupportedTests<armnn::ClWorkloadFactory, armnn::ConvertFp32ToFp16Layer,
-      armnn::DataType::Float16, armnn::DataType::Float16>(reasonIfUnsupported);
-
-    BOOST_CHECK(!result);
-    BOOST_CHECK_EQUAL(reasonIfUnsupported, "Input should be Float32");
-}
-
-BOOST_FIXTURE_TEST_CASE(IsConvertFp32ToFp16SupportedFp32OutputCl, ClContextControlFixture)
-{
-    std::string reasonIfUnsupported;
-
-    bool result = IsConvertLayerSupportedTests<armnn::ClWorkloadFactory, armnn::ConvertFp32ToFp16Layer,
-      armnn::DataType::Float32, armnn::DataType::Float32>(reasonIfUnsupported);
-
-    BOOST_CHECK(!result);
-    BOOST_CHECK_EQUAL(reasonIfUnsupported, "Output should be Float16");
-}
-#endif //#ifdef ARMCOMPUTECL_ENABLED.
-
-BOOST_AUTO_TEST_SUITE_END()
diff --git a/src/backends/test/IsLayerSupportedTestImpl.hpp b/src/backends/test/IsLayerSupportedTestImpl.hpp
index 0f31c8ed08..867f7f2732 100644
--- a/src/backends/test/IsLayerSupportedTestImpl.hpp
+++ b/src/backends/test/IsLayerSupportedTestImpl.hpp
@@ -537,11 +537,6 @@ bool LayerTypeMatchesTestImpl(Tag<Type>)
         LayerTypeMatchesTestImpl<NextType(Type)>(Tag<NextType(Type)>());
 };
 
-bool LayerTypeMatchesTest()
-{
-    return LayerTypeMatchesTestImpl<armnn::LayerType::FirstLayer>(Tag<armnn::LayerType::FirstLayer>());
-};
-
 template<typename FactoryType, typename LayerType, armnn::DataType InputDataType , armnn::DataType OutputDataType>
 bool IsConvertLayerSupportedTests(std::string& reasonIfUnsupported)
 {
diff --git a/src/backends/test/LayerTests.cpp b/src/backends/test/LayerTests.cpp
index 17f3ae12e1..d88db06411 100644
--- a/src/backends/test/LayerTests.cpp
+++ b/src/backends/test/LayerTests.cpp
@@ -39,7 +39,7 @@
 #include "ConvertFp16ToFp32TestImpl.hpp"
 #include "ConvertFp32ToFp16TestImpl.hpp"
 
-#include "ClContextControlFixture.hpp"
+#include <backends/cl/test/ClContextControlFixture.hpp>
 
 // 3-channel 16x8 image used as common input data for a number of Conv2d tests.
 static std::vector<float> ConvInput3x8x16({
diff --git a/src/backends/test/Reference.cpp b/src/backends/test/Reference.cpp
deleted file mode 100644
index 05ebf2e8b0..0000000000
--- a/src/backends/test/Reference.cpp
+++ /dev/null
@@ -1,273 +0,0 @@
-//
-// Copyright © 2017 Arm Ltd. All rights reserved.
-// SPDX-License-Identifier: MIT
-//
-#include <boost/test/unit_test.hpp>
-
-#include "LayerTests.hpp"
-#include "test/TensorHelpers.hpp"
-
-#include <backends/reference/RefWorkloadFactory.hpp>
-
-#include "test/UnitTests.hpp"
-
-BOOST_AUTO_TEST_SUITE(Compute_Reference)
-using FactoryType = armnn::RefWorkloadFactory;
-
-// ============================================================================
-// UNIT tests
-
-// Convolution
-ARMNN_AUTO_TEST_CASE(SimpleConvolution2d3x5, SimpleConvolution2d3x5Test, true)
-ARMNN_AUTO_TEST_CASE(SimpleConvolution2d3x5Uint8, SimpleConvolution2d3x5Uint8Test, true)
-
-ARMNN_AUTO_TEST_CASE(UnbiasedConvolution2d, SimpleConvolution2d3x5Test, false)
-ARMNN_AUTO_TEST_CASE(UnbiasedConvolutionUint8, SimpleConvolution2d3x5Uint8Test, false)
-
-ARMNN_AUTO_TEST_CASE(SimpleConvolution1d, Convolution1dTest, true)
-ARMNN_AUTO_TEST_CASE(SimpleConvolution1dUint8, Convolution1dUint8Test, true)
-
-ARMNN_AUTO_TEST_CASE(SimpleConvolution2d3x3, SimpleConvolution2d3x3Test, true)
-ARMNN_AUTO_TEST_CASE(SimpleConvolution2d3x3Uint8, SimpleConvolution2d3x3Uint8Test, true)
-
-ARMNN_AUTO_TEST_CASE(UnbiasedConvolution2dSquare, SimpleConvolution2d3x3Test, false)
-
-ARMNN_AUTO_TEST_CASE(SimpleConvolution2dAsymmetricPaddingLargerThanHalfKernelSize,
-    Convolution2dAsymmetricPaddingLargerThanHalfKernelSizeTest)
-ARMNN_AUTO_TEST_CASE(SimpleConvolution2dAsymmetricPadding, Convolution2dAsymmetricPaddingTest)
-
-// Depthwise Convolution
-ARMNN_AUTO_TEST_CASE(DepthwiseConvolution2d, DepthwiseConvolution2dTest, true)
-ARMNN_AUTO_TEST_CASE(DepthwiseConvolution2dUint8, DepthwiseConvolution2dUint8Test, true)
-
-ARMNN_AUTO_TEST_CASE(UnbiasedDepthwiseConvolution2d, DepthwiseConvolution2dTest, false)
-ARMNN_AUTO_TEST_CASE(UnbiasedDepthwiseConvolution2dUint8, DepthwiseConvolution2dUint8Test, false)
-
-ARMNN_AUTO_TEST_CASE(DepthwiseConvolution2dDepthMul1, DepthwiseConvolution2dDepthMul1Test, true)
-ARMNN_AUTO_TEST_CASE(DepthwiseConvolution2dDepthMul1Uint8, DepthwiseConvolution2dDepthMul1Uint8Test, true)
-
-ARMNN_AUTO_TEST_CASE(UnbiasedDepthwiseConvolution2dDepthMul1, DepthwiseConvolution2dDepthMul1Test, false)
-ARMNN_AUTO_TEST_CASE(UnbiasedDepthwiseConvolution2dDepthMul1Uint8, DepthwiseConvolution2dDepthMul1Uint8Test, false)
-
-ARMNN_AUTO_TEST_CASE(DepthwiseConvolution2dAsymmetric, DepthwiseConvolution2dAsymmetricTest, true)
-ARMNN_AUTO_TEST_CASE(UnbiasedDepthwiseConvolution2dAsymmetric, DepthwiseConvolution2dAsymmetricTest, false)
-
-// Pooling
-ARMNN_AUTO_TEST_CASE(SimpleMaxPooling2dSize2x2Stride2x2, SimpleMaxPooling2dSize2x2Stride2x2Test, false)
-ARMNN_AUTO_TEST_CASE(SimpleMaxPooling2dSize2x2Stride2x2Uint8, SimpleMaxPooling2dSize2x2Stride2x2Uint8Test, false)
-
-ARMNN_AUTO_TEST_CASE(SimpleMaxPooling2dSize3x3Stride2x4, SimpleMaxPooling2dSize3x3Stride2x4Test, false)
-ARMNN_AUTO_TEST_CASE(SimpleMaxPooling2dSize3x3Stride2x4Uint8, SimpleMaxPooling2dSize3x3Stride2x4Uint8Test, false)
-
-ARMNN_AUTO_TEST_CASE(IgnorePaddingSimpleMaxPooling2d, IgnorePaddingSimpleMaxPooling2dTest)
-ARMNN_AUTO_TEST_CASE(IgnorePaddingSimpleMaxPooling2dUint8, IgnorePaddingSimpleMaxPooling2dUint8Test)
-ARMNN_AUTO_TEST_CASE(IgnorePaddingMaxPooling2dSize3, IgnorePaddingMaxPooling2dSize3Test)
-ARMNN_AUTO_TEST_CASE(IgnorePaddingMaxPooling2dSize3Uint8, IgnorePaddingMaxPooling2dSize3Uint8Test)
-
-ARMNN_AUTO_TEST_CASE(IgnorePaddingSimpleAveragePooling2d, IgnorePaddingSimpleAveragePooling2dTest)
-ARMNN_AUTO_TEST_CASE(IgnorePaddingSimpleAveragePooling2dUint8, IgnorePaddingSimpleAveragePooling2dUint8Test)
-ARMNN_AUTO_TEST_CASE(IgnorePaddingSimpleAveragePooling2dNoPadding, IgnorePaddingSimpleAveragePooling2dNoPaddingTest)
-ARMNN_AUTO_TEST_CASE(IgnorePaddingSimpleAveragePooling2dNoPaddingUint8,
-    IgnorePaddingSimpleAveragePooling2dNoPaddingUint8Test)
-ARMNN_AUTO_TEST_CASE(IgnorePaddingAveragePooling2dSize3, IgnorePaddingAveragePooling2dSize3Test)
-ARMNN_AUTO_TEST_CASE(IgnorePaddingAveragePooling2dSize3Uint8, IgnorePaddingAveragePooling2dSize3Uint8Test)
-
-ARMNN_AUTO_TEST_CASE(IgnorePaddingSimpleL2Pooling2d, IgnorePaddingSimpleL2Pooling2dTest)
-ARMNN_AUTO_TEST_CASE(IgnorePaddingSimpleL2Pooling2dUint8, IgnorePaddingSimpleL2Pooling2dUint8Test)
-ARMNN_AUTO_TEST_CASE(IgnorePaddingL2Pooling2dSize3, IgnorePaddingL2Pooling2dSize3Test)
-ARMNN_AUTO_TEST_CASE(IgnorePaddingL2Pooling2dSize3Uint8, IgnorePaddingL2Pooling2dSize3Uint8Test)
-
-ARMNN_AUTO_TEST_CASE(SimpleAveragePooling2d, SimpleAveragePooling2dTest)
-ARMNN_AUTO_TEST_CASE(SimpleAveragePooling2dUint8, SimpleAveragePooling2dUint8Test)
-ARMNN_AUTO_TEST_CASE(IgnorePaddingAveragePooling2dSize3x2Stride2x2,
-                     IgnorePaddingAveragePooling2dSize3x2Stride2x2Test, false)
-ARMNN_AUTO_TEST_CASE(IgnorePaddingAveragePooling2dSize3x2Stride2x2NoPadding,
-                     IgnorePaddingAveragePooling2dSize3x2Stride2x2Test, true)
-
-ARMNN_AUTO_TEST_CASE(LargeTensorsAveragePooling2d, LargeTensorsAveragePooling2dTest)
-ARMNN_AUTO_TEST_CASE(LargeTensorsAveragePooling2dUint8, LargeTensorsAveragePooling2dUint8Test)
-
-ARMNN_AUTO_TEST_CASE(SimpleL2Pooling2d, SimpleL2Pooling2dTest)
-ARMNN_AUTO_TEST_CASE(SimpleL2Pooling2dUint8, SimpleL2Pooling2dUint8Test)
-
-ARMNN_AUTO_TEST_CASE(L2Pooling2dSize7, L2Pooling2dSize7Test)
-ARMNN_AUTO_TEST_CASE(L2Pooling2dSize7Uint8, L2Pooling2dSize7Uint8Test)
-
-ARMNN_AUTO_TEST_CASE(AsymmNonSquarePooling2d, AsymmetricNonSquarePooling2dTest)
-ARMNN_AUTO_TEST_CASE(AsymmNonSquarePooling2dUint8, AsymmetricNonSquarePooling2dUint8Test)
-
-// Activation
-ARMNN_AUTO_TEST_CASE(ConstantLinearActivation, ConstantLinearActivationTest)
-ARMNN_AUTO_TEST_CASE(ConstantLinearActivationUint8, ConstantLinearActivationUint8Test)
-
-ARMNN_AUTO_TEST_CASE(SimpleNormalizationAcross, SimpleNormalizationAcrossTest)
-ARMNN_AUTO_TEST_CASE(SimpleNormalizationWithin, SimpleNormalizationWithinTest)
-
-ARMNN_AUTO_TEST_CASE(SimpleSoftmaxBeta1, SimpleSoftmaxTest, 1.0f)
-ARMNN_AUTO_TEST_CASE(SimpleSoftmaxBeta2, SimpleSoftmaxTest, 2.0f)
-ARMNN_AUTO_TEST_CASE(SimpleSoftmaxBeta1Uint8, SimpleSoftmaxUint8Test, 1.0f)
-ARMNN_AUTO_TEST_CASE(SimpleSoftmaxBeta2Uint8, SimpleSoftmaxUint8Test, 2.0f)
-
-ARMNN_AUTO_TEST_CASE(SimpleSigmoid, SimpleSigmoidTest)
-ARMNN_AUTO_TEST_CASE(SimpleSigmoidUint8, SimpleSigmoidUint8Test)
-
-ARMNN_AUTO_TEST_CASE(ReLu1, BoundedReLuUpperAndLowerBoundTest)
-ARMNN_AUTO_TEST_CASE(ReLu6, BoundedReLuUpperBoundOnlyTest)
-ARMNN_AUTO_TEST_CASE(ReLu1Uint8, BoundedReLuUint8UpperAndLowerBoundTest)
-ARMNN_AUTO_TEST_CASE(ReLu6Uint8, BoundedReLuUint8UpperBoundOnlyTest)
-
-// Fully Conected
-ARMNN_AUTO_TEST_CASE(SimpleFullyConnected, FullyConnectedFloat32Test, false, false)
-ARMNN_AUTO_TEST_CASE(FullyConnectedUint8, FullyConnectedUint8Test, false)
-ARMNN_AUTO_TEST_CASE(SimpleFullyConnectedWithBias, FullyConnectedFloat32Test, true, false)
-ARMNN_AUTO_TEST_CASE(FullyConnectedBiasedUint8, FullyConnectedUint8Test, true)
-ARMNN_AUTO_TEST_CASE(SimpleFullyConnectedWithTranspose, FullyConnectedFloat32Test, false, true)
-
-ARMNN_AUTO_TEST_CASE(FullyConnectedLarge, FullyConnectedLargeTest, false)
-ARMNN_AUTO_TEST_CASE(FullyConnectedLargeTransposed, FullyConnectedLargeTest, true)
-
-// Splitter
-ARMNN_AUTO_TEST_CASE(SimpleSplitter, SplitterTest)
-ARMNN_AUTO_TEST_CASE(SimpleSplitterUint8, SplitterUint8Test)
-
-ARMNN_AUTO_TEST_CASE(CopyViaSplitter, CopyViaSplitterTest)
-ARMNN_AUTO_TEST_CASE(CopyViaSplitterUint8, CopyViaSplitterUint8Test)
-
-// Merger
-ARMNN_AUTO_TEST_CASE(SimpleMerger, MergerTest)
-ARMNN_AUTO_TEST_CASE(MergerUint8, MergerUint8Test)
-
-// Add
-ARMNN_AUTO_TEST_CASE(SimpleAdd, AdditionTest)
-ARMNN_AUTO_TEST_CASE(AddBroadcast1Element, AdditionBroadcast1ElementTest)
-ARMNN_AUTO_TEST_CASE(AddBroadcast, AdditionBroadcastTest)
-
-ARMNN_AUTO_TEST_CASE(AdditionUint8, AdditionUint8Test)
-ARMNN_AUTO_TEST_CASE(AddBroadcastUint8, AdditionBroadcastUint8Test)
-ARMNN_AUTO_TEST_CASE(AddBroadcast1ElementUint8, AdditionBroadcast1ElementUint8Test)
-
-// Sub
-ARMNN_AUTO_TEST_CASE(SimpleSub, SubtractionTest)
-ARMNN_AUTO_TEST_CASE(SubBroadcast1Element, SubtractionBroadcast1ElementTest)
-ARMNN_AUTO_TEST_CASE(SubBroadcast, SubtractionBroadcastTest)
-
-ARMNN_AUTO_TEST_CASE(SubtractionUint8, SubtractionUint8Test)
-ARMNN_AUTO_TEST_CASE(SubBroadcastUint8, SubtractionBroadcastUint8Test)
-ARMNN_AUTO_TEST_CASE(SubBroadcast1ElementUint8, SubtractionBroadcast1ElementUint8Test)
-
-// Div
-ARMNN_AUTO_TEST_CASE(SimpleDivision, DivisionTest)
-ARMNN_AUTO_TEST_CASE(DivisionByZero, DivisionByZeroTest)
-ARMNN_AUTO_TEST_CASE(DivisionBroadcast1Element, DivisionBroadcast1ElementTest)
-ARMNN_AUTO_TEST_CASE(DivisionBroadcast1DVector, DivisionBroadcast1DVectorTest)
-// NOTE: division by zero for quantized div needs more attention
-//       see IVGCVSW-1849
-ARMNN_AUTO_TEST_CASE(DivisionUint8, DivisionUint8Test)
-ARMNN_AUTO_TEST_CASE(DivisionUint8Broadcast1Element, DivisionBroadcast1ElementUint8Test)
-ARMNN_AUTO_TEST_CASE(DivisionUint8Broadcast1DVector, DivisionBroadcast1DVectorUint8Test)
-
-// Mul
-ARMNN_AUTO_TEST_CASE(SimpleMultiplication, MultiplicationTest)
-ARMNN_AUTO_TEST_CASE(MultiplicationBroadcast1Element, MultiplicationBroadcast1ElementTest)
-ARMNN_AUTO_TEST_CASE(MultiplicationBroadcast1DVector, MultiplicationBroadcast1DVectorTest)
-ARMNN_AUTO_TEST_CASE(MultiplicationUint8, MultiplicationUint8Test)
-ARMNN_AUTO_TEST_CASE(MultiplicationBroadcast1ElementUint8, MultiplicationBroadcast1ElementUint8Test)
-ARMNN_AUTO_TEST_CASE(MultiplicationBroadcast1DVectorUint8, MultiplicationBroadcast1DVectorUint8Test)
-
-// Batch Norm
-ARMNN_AUTO_TEST_CASE(BatchNorm, BatchNormTest)
-ARMNN_AUTO_TEST_CASE(BatchNormUint8, BatchNormUint8Test)
-
-// Resize Bilinear
-ARMNN_AUTO_TEST_CASE(SimpleResizeBilinear, SimpleResizeBilinearTest)
-ARMNN_AUTO_TEST_CASE(SimpleResizeBilinearUint8, SimpleResizeBilinearUint8Test)
-ARMNN_AUTO_TEST_CASE(ResizeBilinearNop, ResizeBilinearNopTest)
-ARMNN_AUTO_TEST_CASE(ResizeBilinearNopUint8, ResizeBilinearNopUint8Test)
-ARMNN_AUTO_TEST_CASE(ResizeBilinearSqMin, ResizeBilinearSqMinTest)
-ARMNN_AUTO_TEST_CASE(ResizeBilinearSqMinUint8, ResizeBilinearSqMinUint8Test)
-ARMNN_AUTO_TEST_CASE(ResizeBilinearMin, ResizeBilinearMinTest)
-ARMNN_AUTO_TEST_CASE(ResizeBilinearMinUint8, ResizeBilinearMinUint8Test)
-ARMNN_AUTO_TEST_CASE(ResizeBilinearMag, ResizeBilinearMagTest)
-ARMNN_AUTO_TEST_CASE(ResizeBilinearMagUint8, ResizeBilinearMagUint8Test)
-
-// Fake Quantization
-ARMNN_AUTO_TEST_CASE(FakeQuantization, FakeQuantizationTest)
-
-// L2 Normalization
-ARMNN_AUTO_TEST_CASE(L2Normalization1d, L2Normalization1dTest)
-ARMNN_AUTO_TEST_CASE(L2Normalization2d, L2Normalization2dTest)
-ARMNN_AUTO_TEST_CASE(L2Normalization3d, L2Normalization3dTest)
-ARMNN_AUTO_TEST_CASE(L2Normalization4d, L2Normalization4dTest)
-
-// NOTE: These tests are disabled until NHWC is supported by the reference L2Normalization implementation.
-//ARMNN_AUTO_TEST_CASE(L2Normalization1dNhwc, L2Normalization1dNhwcTest);
-//ARMNN_AUTO_TEST_CASE(L2Normalization2dNhwc, L2Normalization2dNhwcTest);
-//ARMNN_AUTO_TEST_CASE(L2Normalization3dNhwc, L2Normalization3dNhwcTest);
-//ARMNN_AUTO_TEST_CASE(L2Normalization4dNhwc, L2Normalization4dNhwcTest);
-
-// Constant
-ARMNN_AUTO_TEST_CASE(Constant, ConstantTest)
-ARMNN_AUTO_TEST_CASE(ConstantUint8, ConstantUint8Test)
-
-// Concat
-ARMNN_AUTO_TEST_CASE(Concatenation1d, Concatenation1dTest)
-ARMNN_AUTO_TEST_CASE(Concatenation1dUint8, Concatenation1dUint8Test)
-
-ARMNN_AUTO_TEST_CASE(Concatenation2dDim0, Concatenation2dDim0Test)
-ARMNN_AUTO_TEST_CASE(Concatenation2dDim0Uint8, Concatenation2dDim0Uint8Test)
-ARMNN_AUTO_TEST_CASE(Concatenation2dDim1, Concatenation2dDim1Test)
-ARMNN_AUTO_TEST_CASE(Concatenation2dDim1Uint8, Concatenation2dDim1Uint8Test)
-
-ARMNN_AUTO_TEST_CASE(Concatenation2dDim0DiffInputDims, Concatenation2dDim0DiffInputDimsTest)
-ARMNN_AUTO_TEST_CASE(Concatenation2dDim0DiffInputDimsUint8, Concatenation2dDim0DiffInputDimsUint8Test)
-ARMNN_AUTO_TEST_CASE(Concatenation2dDim1DiffInputDims, Concatenation2dDim1DiffInputDimsTest)
-ARMNN_AUTO_TEST_CASE(Concatenation2dDim1DiffInputDimsUint8, Concatenation2dDim1DiffInputDimsUint8Test)
-
-ARMNN_AUTO_TEST_CASE(Concatenation3dDim0, Concatenation3dDim0Test)
-ARMNN_AUTO_TEST_CASE(Concatenation3dDim0Uint8, Concatenation3dDim0Uint8Test)
-ARMNN_AUTO_TEST_CASE(Concatenation3dDim1, Concatenation3dDim1Test)
-ARMNN_AUTO_TEST_CASE(Concatenation3dDim1Uint8, Concatenation3dDim1Uint8Test)
-ARMNN_AUTO_TEST_CASE(Concatenation3dDim2, Concatenation3dDim2Test)
-ARMNN_AUTO_TEST_CASE(Concatenation3dDim2Uint8, Concatenation3dDim2Uint8Test)
-
-ARMNN_AUTO_TEST_CASE(Concatenation3dDim0DiffInputDims, Concatenation3dDim0DiffInputDimsTest)
-ARMNN_AUTO_TEST_CASE(Concatenation3dDim0DiffInputDimsUint8, Concatenation3dDim0DiffInputDimsUint8Test)
-ARMNN_AUTO_TEST_CASE(Concatenation3dDim1DiffInputDims, Concatenation3dDim1DiffInputDimsTest)
-ARMNN_AUTO_TEST_CASE(Concatenation3dDim1DiffInputDimsUint8, Concatenation3dDim1DiffInputDimsUint8Test)
-ARMNN_AUTO_TEST_CASE(Concatenation3dDim2DiffInputDims, Concatenation3dDim2DiffInputDimsTest)
-ARMNN_AUTO_TEST_CASE(Concatenation3dDim2DiffInputDimsUint8, Concatenation3dDim2DiffInputDimsUint8Test)
-
-// Floor
-ARMNN_AUTO_TEST_CASE(SimpleFloor, SimpleFloorTest)
-
-// Reshape
-ARMNN_AUTO_TEST_CASE(SimpleReshapeFloat32, SimpleReshapeFloat32Test)
-ARMNN_AUTO_TEST_CASE(SimpleReshapeUint8, SimpleReshapeUint8Test)
-
-// Permute
-ARMNN_AUTO_TEST_CASE(SimplePermuteFloat32, SimplePermuteFloat32Test)
-ARMNN_AUTO_TEST_CASE(SimplePermuteUint8, SimplePermuteUint8Test)
-ARMNN_AUTO_TEST_CASE(PermuteFloat32ValueSet1, PermuteFloat32ValueSet1Test)
-ARMNN_AUTO_TEST_CASE(PermuteFloat32ValueSet2, PermuteFloat32ValueSet2Test)
-ARMNN_AUTO_TEST_CASE(PermuteFloat32ValueSet3, PermuteFloat32ValueSet3Test)
-
-// Convert from Float16 to Float32
-ARMNN_AUTO_TEST_CASE(SimpleConvertFp16ToFp32, SimpleConvertFp16ToFp32Test)
-// Convert from Float32 to Float16
-ARMNN_AUTO_TEST_CASE(SimpleConvertFp32ToFp16, SimpleConvertFp32ToFp16Test)
-
-// Mean
-ARMNN_AUTO_TEST_CASE(MeanUint8Simple, MeanUint8SimpleTest)
-ARMNN_AUTO_TEST_CASE(MeanUint8SimpleAxis, MeanUint8SimpleAxisTest)
-ARMNN_AUTO_TEST_CASE(MeanUint8KeepDims, MeanUint8KeepDimsTest)
-ARMNN_AUTO_TEST_CASE(MeanUint8MultipleDims, MeanUint8MultipleDimsTest)
-ARMNN_AUTO_TEST_CASE(MeanVtsUint8, MeanVtsUint8Test)
-
-ARMNN_AUTO_TEST_CASE(MeanFloatSimple, MeanFloatSimpleTest)
-ARMNN_AUTO_TEST_CASE(MeanFloatSimpleAxis, MeanFloatSimpleAxisTest)
-ARMNN_AUTO_TEST_CASE(MeanFloatKeepDims, MeanFloatKeepDimsTest)
-ARMNN_AUTO_TEST_CASE(MeanFloatMultipleDims, MeanFloatMultipleDimsTest)
-ARMNN_AUTO_TEST_CASE(MeanVtsFloat1, MeanVtsFloat1Test)
-ARMNN_AUTO_TEST_CASE(MeanVtsFloat2, MeanVtsFloat2Test)
-
-BOOST_AUTO_TEST_SUITE_END()