The supermassive black holes (SMBHs; 10^6-10M⊙) sitting at the centre of most massive galaxies act as enormous reservoirs of energy, accreting gas from the galaxy and efficiently converting part of the energy into radiation, outflows and jets. Studying this fundamental process involved in the growth of the central BHs and their feedback into their host galaxies is crucial to understand how BHs shaped the large-scale structures of the Universe as we observe them today. Surprisingly, the deepest infrared surveys have revealed that SMBHs with masses 10^8-9M⊙ were already present when the Universe was only 5% of its current age, suggesting that these BHs must have grown extremely fast, but the extreme accretion regime that would be required to allow these BHs to grow so fast is poorly understood. While studying these early SMBHs in detail is not feasible with current instrumentation, after two decades of studies, there is now firm evidence supporting that most of the Ultraluminous X-ray sources (ULXs) typically found in external galaxies are accreting binary systems undergoing sustained super-Eddington accretion. Much like a scaled-down version of a SMBH, the copious amounts of radiation and outflows these mighty sources produce can ionise large nebulae of gas extending hundreds of parsecs in size, offering an observational template to study the feedback and efficiency of the super-Eddington regime. Yet the nature of the accretor in most of them remains unknown, while a few show X-ray pulsations indicative of a neutron star (NS) accretor. Moreover, the accretion flow geometry allowing them to reach such extreme luminosities and its dependency with the nature of the accretor is still highly debated. This thesis presents a multi-wavelength study in an effort to shed light on the engines behind ULXs, the properties of their inflows and outflows as well the feedback signatures in their surroundings. Two main lines of investigation are presented. In the first one, I exploited the extensive archival data offered by the major X-ray telescopes (XMM-Newton, Chandra and NuSTAR) to present a comprehensive study of the long-term spectral transitions of a broad sample of high-quality data ULXs, including known pulsating NS-ULXs, in order to test which of the current models to explain their extreme luminosities fit the data best. The analysis suggest that for the most part, their spectral transitions are best described by models invoking anisotropic emission induced by super-Eddington winds, with accreting lowly magnetised (B < 10^12G) NSs or BHs as their engines. The wealth of data served also to examine the spectral differences/similarities between the known NS-ULXs and those for which the accretor is unknown, revealing a potential new NS-ULX. This work is followed by an in-depth study of two archetypal ULXs, Holmberg II X-1 and NGC 5204 X-1, where I exploited the high-observing cadence of Swift-XRT to reveal that these two sources follow a spectral recurrent evolutionary cycle. The spectral transitions and variability associated with the cycle is discussed in the framework of super-Eddington accretion, by invoking mass-accretion rate changes and varying degree of obscuration by the super-Eddington winds. [...]