Text this: Compressing Large-Scale Transformer-Based Models: A Case Study on BERT