Recommender Systems (RS) are nowadays widely used in the area of consumer electronics and home entertainment. They are exploited by large companies (such as Amazon in the area of e-commerce) and used by millions of users (e.g. 93 million for Netflix). With Recommender Systems, users explore possible items of interest, consult details about these items and read explanations about the choices offered, but these tasks are relevant to other application domains too. Recommender Systems can be a powerful option to support operators of critical systems who can be confronted with choosing the right option from a set of alternative (especially in the case of incidents). However, deploying recommender systems in critical contexts requires applying critical systems engineering practices throughout its development. As recommender systems are a special kind of interactive systems, those engineering practices must integrate and reconcile the ones of critical systems engineering and interactive systems engineering. This article presents a comprehensive study of the state of the art in recommender systems engineering which highlights the fact that their engineering is still in its infancy. In the light of Recommender Systems characteristics, we propose a systematic analysis of standards in critical systems engineering and knowledge from dependable computing field to extract a set of requirements that are relevant to Recommender Systems. As those requirements target at different components of a systems, we first propose a generic architecture to decompose Recommender Systems. This generic architecture integrates existing proposals from the Recommender Systems community with current knowledge in interactive systems architectures. In order to engineer Recommender Systems compliant with the entire list of requirements identified, we propose to use a set of complementary integrated model-based approaches from the literature. This approach is illustrated on a large case study about future aircraft alerting systems, which highlights the potential benefits of the usage of recommender systems in critical contexts.