Advances in distributed systems and Internet technologies have prompted the emergence of new applications of computer systems including computer grids, ubiquitous computing, and business integration systems. Conventional development methods like object-oriented and component development cannot be applied in volatile and globally dispersed Internet environment. Thus, the main development model in service-oriented systems is service composition. Service composition process is based on combining and connecting a set of services in order to create new composite services and applications. The existing service composition systems are based on centralized control architecture that reduces the reliability and scalability of the distributed system. The doctoral thesis proposes and describes the service-oriented programming model, which is a new development model designed to enable development of fully distributed computer systems. The basic principles of the serviceoriented programming model are virtualization of computer resources as Web Services, twolayered management of service deployment, three-tiered logical architecture, distributed control and management of system execution, and multi-layered application deployment process. The original contribution of the proposed programming model is the three-tiered logical architecture consisting of a set of application services that implement processing logic, a set of communication and synchronization services that interconnect the distributed system components, and a set of distributed programs that integrate application services by leveraging the communication and synchronization services. The thesis also describes the PIE development environment, which enables the implementation of service-oriented programming model. A set of measurements has been performed in order to test the performance of the implemented service-oriented programming model by varying various application parameters. Obtained measurement results indicate that the distributed model of control and management in complex distributed systems is significantly more efficient than the centralized models.