To address the challenge of real-time monitoring and control of cladding layer dimensions during laser directed energy deposition (LDED), we propose an integrated framework combining an enhanced CondenseNet architecture with gated recurrent units (GRUs) for online measurement and prediction. The framework comprises two key components: A modified CondenseNet algorithm that fuses key process parameters and melt-pool images to achieve realtime measurement of cladding layer dimensions; A temporal modeling module based on GRUs, which utilizes historical dimension sequences to predict future cladding layer height. Experimental results demonstrate an average percentage error of 5.68% for width measurements and 3.72% for height measurements, with an inference time of 17.6 ms per image under computational resource constraints. Leveraging these real-time measurements, the GRU-based predictor further achieves a height prediction error of 3.60%. The proposed framework enables high-precision, real-time monitoring of cladding layer dimensions with limited computational resources, offering a robust solution for closed-loop control in LDED processes.