Considering process, voltage and temperature (PVT) corners during the automatic synthesis of millimeter-wavelength (mmWave) integrated circuits (ICs) in nanometer technologies is incurring prohibitive optimization times. Therefore, this paper presents innovative research towards the automation of mmWave IC design by using deep learning (DL) to assist a simulation-based sizing tool, via independent and parallel PVT performance regressors that bypass the simulator. Transfer learning from nominal (TT) to corner conditions avoids the need for expensive PVT data, and an online Bayesian-assisted incremental learning is used to refine each model with accurate simulator data. The proposed methodology is tested on the design two low-noise amplifiers (LNAs) in the challenging 28-GHz band for a 65-nm node. When compared with full simulation-based synthesis, it reduced the workload of the circuit simulator up to ~69% while achieving a speed-up factor of ~3.4×, and, requiring less ~86% dataset generation effort when compared with most recent PVT regressors.