Models of the origin of spheroidal stellar systems, or cluster formation scenarios, need to account for empirical correlations both between scale and velocity dispersion σ and between luminosity and σ found in star-forming regions and relaxed spheroidal stellar systems. The model here proposed accounts for both correlations if the stellar system formation follows a particular sequence. This requires that the quasi-static collapse of a protocluster cloud be halted as soon as stars begin to form, and this occurs once fragments acquire stellar sizes, if the fragment temperature remains at a constant value of about 10 K. The collection of pre-mainsequence low-mass stars undergoing winds while moving with a velocity dispersion σ* will soon stir the remaining cloud, providing it with an average turbulent motion σgas σ*. The cloud agitation is here proposed to be caused by the endless supersonic passage of isothermal bow shocks, or "cometary" shocks, generated by the stellar wind sources ramming through the leftover cloud. These maintain supersonic turbulence and lead also to a distinct structure of the remaining cloud. This mechanism leads to an estimate of the total wind power required and the corresponding cluster luminosity. The latter agrees with observations both in general magnitude and in its correlation with velocity dispersion (Lcluster σ σ4). Following stability, star formation continues, at least at the rate needed to keep the cloud from collapsing any further until the birth of massive stars, which by means of photoionization heat up the remaining matter and inhibit any further star formation, and thus mark the end of cluster formation. H II regions produced by massive clusters will display broad lines reflecting the supersonic σgas acquired from the cometary passage of the wind-driven sources. Afterward, the supersonic H II region expansion, and/or any further localized major input of energy, such as supernova explosions, will rapidly lead to larger velocities and to the removal of gas from the starr-forming region, causing broader but lower intensity emission lines. The model is confronted with recent data on giant H II regions showing excellent qualitative and quantitative agreement.