Rjava performance


We are invoking machine learning R models like “SVM”, “Navie Byes” from java. Java - R communication we are using RJava. Problem here is models require huge training data set. So every time model is invoked from java, training data set (which is quite huge) for instance
model <- naiveBayes(as.factor(RAG.Status)~., data=data) also is invoked. since training data set hardly changes, is it possible using (rjava/rserve)for model to be cached and only prediction happens for test data which changes?