The recent deployment of 5G networks is bringing benefits to the population but it is also raising public concern about human RF-EMF exposure levels. This is particularly relevant considering the next 5G mobile devices, which are placed in close proximity to the subjects. Therefore, the aim of the following paper is focused on expanding the knowledge of the exposure levels in 5G exposure scenarios, specifically for mobile applications, using computational methods. The mobile antenna was designed considering the 5G technology innovations (i.e., mm-wave spectrum, beamforming capability, high gain and wide coverage), resulting in a phased-array antenna with eight elements at the working frequency of 27 GHz. To assess the exposure levels, different types of skin models with different grades of details and layers were considered. Furthermore, not only was the presence of a mobile phone user simulated, but also that of a person in their proximity, who could be hit by the main beam of the phased-array antenna. All the simulations were conducted in Sim4Life platform, where the exposure levels were assessed in terms of absorbed power density averaged over 4 cm and 1 cm, following the ICNIRP guidelines. The results highlighted that the use of the homogeneous skin model led to the absorbed power density peaks being greatly underestimated, with respect to those obtained in multilayer skin models. Furthermore, interestingly, we found that the exposure levels obtained for the person passing nearby were slightly higher than those experienced by the mobile phone user himself. Finally, using the allowed input power for real mobile applications, all the values remained below the limits indicated by the ICNIRP guidelines.