Towards Building Efficient Sentence BERT Models using Layer Pruning