Scalable Neural Network Compression and Pruning Using Hard Clustering and L1 Regularization