Compressing Language Models for Specialized Domains