Refactoring is the process of transforming a software system to improve its overall structure while preserving its observable behaviour. Refactoring engines are normally used to perform these transformations for efficiency and in order to avoid introducing behavioural changes into the program due to human error. Although these engines do not verify that behaviour is preserved, it is widely accepted that automated transformations are less likely to introduce errors in comparison to manual refactoring. Despite the advantages provided by refactoring engines they fall foul of certain weaknesses. Here we hypothesise that Model Driven Engineering can be used to produce improved refactoring engines that are less vulnerable to those weaknesses. We develop a Domain Specific Transformation Language for defining new composite refactorings from a set of built–in primitives and to script their application. We also develop an interpreter for the language, effectively providing an operational semantics, in the guise of an extensible transformation framework. We evaluate our approach with a case study examining the correlation between actual and predicted measurements of the Coupling Between Objects metric for classes that undergo the extract class refactoring. The results show that our approach is promising.