Diverse authoring approaches and tools have been designed to assist the creation of units of learning compliant to current learning technology specifications. Although visual and pattern-based editors of Learning Designs (LD) can help to abstract the learning designer from the details of the specifications, they are still far from a high-level, integrated authoring environment. This paper analyzes the major approaches used to transform an abstract LD into a concrete unit of learning (UoL), according to three desired features: the use of patterns and other design techniques to abstract the specific representational details; the difference between the abstract source LD model and the concrete target UoL model; and the possibility of combining multiple models into a single environment. A classification is proposed for the LD techniques commonly found in the analyzed approaches, in order to underline its abstraction from the details of the underlying specifications. We have integrated such LD techniques in a unified Model-Driven Learning Design (MDLD) meta-modelling environment, which has been used to generate UoLs from a number of meta-models. The model-driven development process was studied on the creation of a IMS LD UoL for the Learning Networks' knowledge base.