The knowledge, embodied in machine learning models for intelligent systems, is commonly associated with time-consuming and costly processes such as large-scale data collection, data labelling, network training, and fine-tuning of models. Sharing and reuse of these elaborated models between intelligent systems deployed in a different environment, which is known as transfer learning, would facilitate the adoption of services for the users and accelerates the uptake of intelligent systems in environments such as smart building and smart city applications. In this context, the communication and knowledge exchange between AI-enabled environments depend on a complicated networks of systems, system of systems, digital assets, and their chain of dependencies that hardly follows the centralized schema of traditional information systems. Rather, it requires an adaptive decentralized system architecture that is empowered by features such as data provenance, workflow transparency, and validation of process participants. In this research, we propose a decentralized and adaptive software framework based on blockchain and knowledge graph technologies that supports the knowledge exchange and interoperability between IoT-enabled environments, in a transparent and trustworthy way.