Hardware-Aware Compression Techniques for Embedded Deep Neural Networks