Reference models, being conceptual models created to provide so-called best practices, abstract away from one specific organization, and instead, focus on characteristics common to many organizations, within or across one or more industries and/or application domains. Although reference models come with several promises (e.g., fostering reuse and a shared domain understanding), the design and adoption of reference models face many challenges. These challenges include finding a balance between generality and specificity, supporting variability, and consistent adaptation of a reference model. We show in our research that the mentioned reference modeling challenges are partially related to the characteristics of the modeling languages used to create and disseminate reference models. A multi-level modeling language architecture offers expressiveness and flexibility that naturally fit with the idea of reference models, by capitalizing on mechanisms such as a relaxed type/instance dichotomy, or deferred instantiation. Therefore, in our investigation, among others, we apply a multi-level language architecture to reconstruct selected reference models, and compare a reference model as created with conventional meta modeling, with the same reference model as created through multi-level modeling. In our future work, we intend to extend well-established existing methods for reference model construction, with mechanisms that consider the particulars of multi-level modeling.