This paper describes Coq libraries devoted to the semantic of relaxed memory models. These libraries formalise a framework which covers a large class of industrial models. Implementing this framework inside a proof assistant has significantly helped improving its design and crafting the most concise and relevant specifications. Similarly the use of a proof assistant has been instrumental in the study of the semantic of synchronisation primitives, which we illustrate by the formal proof of a barrier placement theorem. We explain the choices we made to re-design our Coq libraries, and in particular what we gained from adopting a small-scale reflection methodology.