The modularized code encapsulating object interactions is characterized by class hierarchies. In the implementation of mobile agents, we have observed that the changes in agent interaction protocols lead to uncontrolled subclassing and consequently to disorder. This phenomenon is known as entropy. The additional subclassing, modification to protocols, restructuring of the class hierarchies, changes to visibility of attributes, and method overloading result in increased co mplexity of the code. This problem in agent design has been tackled by Kendall (Kendall, 1999) who proposed development using Aspect/J and separation of concerns. Since there has been no proof of reduced complexity, we have proposed metrics for software complexity estimation, and ranking of compositional elements developed with As-pect/J. The metrics have been tested on Java code for mobile agents.