Nowadays, highly-configurable software systems are widely used throughout all areas of human activity. Modern self-adaptive systems, such as self-organizing networks or digital twins, require wide variability management capabilities to reflect their response to external changes. A typical software architecture for such a highly-configurable system is a software product line, which can be informally described as a platform with static and variable software components. In this paper, we define requirements for variability management of highly-configurable software systems using the example of BRISE, a framework for expensive black-box optimization. Two of these requirements are not supported by state-of-the- art approaches: cross-tree constraints with cardinalities and constraint- based staged configuration and validation mechanisms. Therefore, we present Waffle, a new feature modeling language and corresponding tool that aims to capture the variability of highly-configurable software systems. We evaluated Waffle by creating a feature model for the BRISE framework. Using this feature model, we established an adapted BRISE configuration that found a near-optimal solution for intra-cell user distribution with a reduced number of measured cell transmission power configurations up to 99.89%. Such results are essential to the real- time functioning of digital twins of massive systems with complex underlying processes and dependencies, like cellular networks or a countrywide highway network.