By automation of their critical systems, modern buildings are becoming increasingly intelligent, but also increasingly vulnerable to both cyber and physical attacks. We propose that multi-models can be used not only to assess the security weaknesses of smart buildings, but also to optimise their control to be resilient to malicious use. The proposed approach makes use of the INTO-CPS toolchain to model both building systems and the behaviour of adversaries, and utilises design space exploration to analyse the impact of security on usability. By separation of standard control and security monitoring, the approach is suitable for both the design of new controllers and the improvement of legacy systems. A case study of a fan coil unit demonstrates how a controller can be augmented to be more secure, and how the trade-off between security and usability can be explored to find an optimal design. We propose that the suggested use of multi-models can aid building managers and security engineers to build systems which are both secure and user friendly.