This paper introduces CogNet architecture which is developed based on cross-layer optimized network architecture and specially designed for cognitive radio networks. It enables cognitive radios to share the network information between the lower three layers through a common database while efficiently processing the shared information using the cognitive engine which is attached onto the common database. Cognitive engine in the proposed CogNet architecture is primarily used for routing function in network layer and is served as an example of use of CogNet architecture. It contains four estimators for different purposes and a five-step sequential procedure is implemented to process the shared network information. The available parameters for routing functions, such as routing metrics, can be intelligently adjusted according to the cross-layer optimized feedback from cognitive engine. The detailed modeling, analysis and implementations of CogNet architecture are provided.