CPLAY is a novel serious-game platform combining miniaturized, instrumented building bricks with a tablet interface and AI-driven activity recognition to deliver personalized, gamified upper-limb rehabilitation for children with Developmental Coordination Disorder or Cerebral Palsy. Developed through a co-design process with therapists, engineers, parents and children, two IMU-equipped brick form factors transmit 70 Hz motion data via Bluetooth to a tablet hosting fifteen progressively challenging 3D assembly tasks enriched with rewards and audiovisual feedback. In a clinical feasibility trial, 30 children (mean age 9.1 ± 1.0 years) performed each model twice under supervision; synchronized motion and video data were annotated to train signal-processing routines and deep-learning classifiers. Signal processing detected all 36 movement and fall events (one false positive), while the Transformer model outperformed other approaches, achieving 86.0% multi-class accuracy and 85–96% binary accuracy across activity labels. These results demonstrate CPLAY's ability to objectively detect and classify key rehabilitation movements, paving the way for adaptive, AI-driven therapy, with future work focused on embedding a recommendation engine, scaling trials and refining AI models.