The representation of the problem's parameters is the important issue in the research on genetic algorithms. In this thesis, we developed our own representation scheme, called "Object-Oriented Trucking" approach. Under the "Object-Oriented Trucking" approach each organism is encoded by a fixed-length string. Each entry in the string represents a C++ base class responsible for a particular type of organism's behavior. The value in the string entry selects one of the several programmer-written subclasses or strategies derived from the base class. Each organism has the same set of base classes, which guarantees that all organisms have the same and meaningful structure. However, the strategies used change from organism to organism depending on the values of the string entries. Since the structure of any given organism is not random but rather meaningful, and the programmer-written subclasses used in the organism are also supposed to be meaningful, we can guarantee that any organism would behave meaningfully. Though our approach is proofed against producing nonsense organisms, in some degree it lacks flexible. To solve this problem we propose a so called tree-level system which is able to automatically generate a large number of strategies from a set of primitive functions