Efficient Models for the Detection of Hate, Abuse and Profanity