Minimizing Computational Resources for Deep Machine Learning: A Compression and Neural Architecture Search Perspective for Image Classification and Object Detection