Taking advantage of IoT, services and Cloud technologies , the development of Cloud of Things (CoT) changes the way control applications are engineered and developed. CoT helps to move from a dedicated design and development of control applications to a Control as a Service vision, relying on Cyber Physical System (CPS) composition and orchestration. This vision requires developing a new architecture to provide a loosely coupled way to interconnect objects and control software , mixing both logical and physical constraints. To fit this challenge, we propose a Control as a Service model, extending the Cloud XaaS model to automation. To this end, we design a Control Service, gathering both functional and non functional specifications to describe each component of the system. By this way, control applications can be implemented by selecting and composing these control services associated to the different sensors, controllers and actuators. Fitting the loosely coupled principle of the service oriented architecture, we introduce a data manager component used to store the information produced and exchanged by control components (sensors and controllers) and an event manager, in charge of invoking the services when all the necessary input information are available. A prototype hosting a simulated smart city control application, distributed on different machines, is presented.