Numerical aperture is n sin θ: the refractive index of the medium between lens and specimen, times the sine of the half-angle of the cone of light the lens can accept. It measures how much of the light leaving a point the objective actually catches. A point source radiates in every direction; a lens catches a cone of it; NA is the size of that cone.
Now look at what constrains it. Sine cannot exceed 1. Widen the cone all you like — perfect the glass, spend anything — and in air the NA cannot reach 1.0, with 0.95 about the practical ceiling. This is not an engineering limit that better manufacturing will move. It is trigonometry.
Which leaves exactly one way past the wall, and it is the other term. Raise n. Water gives 1.33, glycerol 1.47, immersion oil 1.515, and suddenly NA above 1 is available — 1.4 routinely, 1.49 for the demanding. Immersion media are not a refinement for people who want a slightly nicer image. They are the only escape from a hard mathematical ceiling, and that is why anyone tolerates the mess.
Think of the lens as catching light the way a bucket catches rain: the wider its mouth, the more it gets. Numerical aperture is the width of that mouth.
But there is a catch built into the arithmetic. However wide you make the opening, in air the number cannot reach 1.0 — not with better glass, not at any price. The only way through is to change what sits between the lens and the sample. Put oil there instead of air and the ceiling lifts. That is the entire reason immersion lenses exist, and why people put up with the fuss of them.