Automatic License Plate Recognition (ALPR) can automate residential vehicle access, but conventional cloud centric and edge only approaches face limitations in latency, privacy, computational capability, and recognition robustness. This study presents a deployed residential smart gate prototype that integrates a confidence aware hybrid ALPR pipeline, a Vehicle Behavior Profiling System (VBPS), and security hardened MQTT/TLS communication on a Raspberry Pi 5. The ALPR pipeline employs YOLOv5n ONNX for license plate detection and dynamically routes detected plates to an in-memory resident cache, local Tesseract OCR, or cloud OCR based on recognition confidence and recent vehicle history, while the VBPS maintains vehicle profiles and evaluates temporal and frequency based behavioural features to generate five adaptive gate decisions instead of a binary allow/deny response. Gate control and monitoring communications are secured using MQTT over TLS 1.2 with AES-256-GCM encryption and mutual certificate authentication. Experimental evaluation of the deployed prototype achieved 97.3% detection mAP, a 61% cache hit rate, a 96.7% reduction in cloud API usage, and a weighted average processing latency of 749 ms. The false denial rate decreased from 51.3% in the evaluated baseline configurations to 3.6% with the behaviour aware system, while the VBPS achieved 100% precision across the evaluated structured temporal and behavioral anomaly scenarios and satisfied all 12 evaluated access control requirements. Furthermore, four security tests covering unauthorized clients, message interception, replay attacks, and command integrity were successfully passed. These findings demonstrate the feasibility of combining hybrid ALPR, behaviour aware access control, and secure IoT communication on constrained residential hardware, while identifying the need for larger real-world datasets and further optimization of end-to-end latency.