Early performance evaluation is essential when designing systems in order to enable decision making. This requires both a way to simulate the system in an early state of design and a set of relevant Key Performance Indicators (KPIs). Model-Driven Engineering and Domain-Specific Languages (DSLs) are well suited for this endeavor, e.g. using executable DSLs fitting for early simulation. However, KPIs are commonly tailored to a particular system, and therefore need to be redefined for each of its variation. In light of these problems, this paper examines how KPIs can be defined directly at the level of a DSL, thus making them available for domain experts at the model level. We demonstrate this idea through a case study centered on a DSL to define, simulate, and evaluate the performance of simple manufacturing systems. Models simulation is performed by the DSL operational semantics, and yields execution traces that can then be analyzed by KPIs defined at the DSL level. Performance results are captured using the Structured Metrics Meta-model. We illustrate the usefulness of the proposed approach and KPIs to evaluate a simple hammer factory model and its subsequent reconfiguration.