In this paper, we discuss the application of search-based software testing techniques for unit level testing of a real-world telecommunication middleware at Ericsson. Input data for the system under test consists of nested data structures, and includes non-trivial variables such as uninitialized pointers. Our current implementation analyzes the existing test cases to discover how to handle pointers, set global system parameters, and any other setup code that needs to run before the actual test case. Hill climbing (HC) and (1+1) evolutionary algorithm (EA) metaheuristic search algorithms are used to generate input data for branch coverage. We compare HC, (1+1)EA, and random search as a baseline of performance with respect to effectiveness, measured as branch coverage, and efficiency, measured as number of executions needed. Difficulties arising from the specialized execution environment and the adaptations for handling these problems are also discussed.