Towards Secure and Private AI: A Framework for Decentralized Inference