On Importance of Pruning and Distillation for Efficient Low Resource NLP