Custom Stemmer

I used elasticsearch default turkish stemmer but I couldn't get enough performance to analyze my data. Is there any way that I can implement my own stemmer algorithm to elasticsearch?