Additive manufacturing (AM) of metals is a complex process to monitor in-situ, as the layer-by-layer deposition and material-beam interactions present a number of challenges. However, a design-driven build of customised and nearly net-shape parts makes it favourable for the manufacture of complex geometries. For process certification of critical parts there is a need for reliable process monitoring and control. Advanced thermal imaging methods can provide information in-situ, this can be used for quality assurance. Existing state-of-the-art studies based on thermal image acquisitions have the limitation of being demonstrated on simple part designs, often symmetric thin walls or cuboid structures. Statistical Process Control (SPC) has been demonstrated in past work as effective in AM, however on simple part geometries. In this work we introduce a multi-model and self-tuning computational framework for SPC via multilinear principal component analysis (MPCA), to address AM process monitoring of geometrically complex parts. In the proposed computational method, process behaviours are expressed via extracting and grouping meltpool features, thus accounting for multiple possible meltpool behaviours corresponding to part complexity and different design features. The framework oper-ates on an iterative fashion, where the clusters (hence captured behaviours) are updated in-situ on a per-layer basis, hence continuously tuning the monitoring algorithm. A case study in blown-powder laser melting deposition of a complex geometry is presented, which includes two manufactured parts where the correlation between the predicted outliers and measured part defects is demonstrated.