Automated software synthesis is one of the primary methods used in knowledge-based software engineering. Although many software synthesis systems have been successfully designed and implemented, automatic control of these systems remains a difficult problem. This paper describes a task-specific framework called MetaMorphos that allows explicit representations of control knowledge about the programming task and contains a selection method based on a given set of features. We apply MetaMorphos in the domain of schema-based software synthesis systems and illustrate how we use MetaMorphos to select data structure and algorithm schemas. An example shows how MetaMorphos handles the difficult problem of selecting schemas for two very similar problems which, in the best case, require different solutions.