We investigate the problem arising in allocating master and workers in the Master-Worker Paradigm. Although various methods have been proposed for Master-Worker allocation, optimal allocation is yet to be achieved. This paper proposes an extension to the generic Master-Worker architecture for achieving optimal allocation of masters and workers. The architecture is extended by introducing an additional utilizing layer in between the Grid application layer and the Grid service layer, which is called Master-worker optimization layer. The layer consists of a knowledge base and uses statistical knowledge for utilization of the resource allocation. The salient feature of this layer is that it uses the prior knowledge and numerical data relating to groups of individual computers for the decision making process. The paper describes the architecture of the extended Master-Worker architecture; its functionality, possible alternatives for the Master-worker optimization layer and assessment of the proposed method.