Metadata describing common Java classes is extracted and utilized to optimize a Java program's performance based on the code requirements of the application. Data mining techniques extract metadata and other information about different sets of JVM byte code written by multiple vendors to determine the design choices and optimizations made by each vendor in implementing their JVM. Then, byte code in a Java application being programmed is analyzed in comparison to each vendor's JVM metadata and information. Based upon the application's requirements (for example footprint, performance, etc.), recommendations are generated that yield a newly compiled rt.jar that combines the best practices of each vendor together or recommends a particular vendor's standard JVM that is best suited for the application.